This week: the most famous AI hedge fund on Wall Street lost two-thirds of its money in a single month. Anthropic admitted its Claude models hacked three real companies during safety tests, and the oldest one kept going even after it realised what it was doing. And OpenAI’s next model solved ten maths problems that had been open for over a decade, for about $2,000 in compute. One of these stories is a lesson. One is a warning. One is a wonder. Let’s sort which is which.

🔷 The big ones

1. The AI hedge fund that bet everything just lost most of it

There’s a 24-year-old named Leopold Aschenbrenner who, two years ago, published a 165-page essay called “Situational Awareness.” It argued that artificial general intelligence could arrive within years and reshape the global economy. The essay went viral in Silicon Valley. Aschenbrenner, a former OpenAI researcher with no investment experience, used the attention to launch a hedge fund with the same name.

The idea was simple: AI is going to need chips, data centres, power, and cloud computing. Buy the companies that supply those things, bet against the companies AI will make obsolete, and ride the wave.

It worked. Spectacularly. The fund grew to $45 billion. It was up 439% in the first half of this year. Aschenbrenner became one of the most closely watched investors on Wall Street. People scrutinised his quarterly filings for clues about the next AI trade.

Then July happened.

AI infrastructure stocks cratered. SK Hynix, Sandisk, Micron, CoreWeave. All down more than 35% in the month. The fund’s long bets collapsed. Its short bets, against software companies like Adobe, went the wrong way too. Both sides of the portfolio lost money at the same time.

The fund had borrowed billions to amplify its bets. When the value of those bets fell, the banks that lent the money started calling it back. Margin calls. The fund had to sell. Selling pushed prices lower. Lower prices meant more margin calls. A death spiral.

On Wednesday night, Wall Street banks held an overnight auction for the fund’s entire public stock portfolio. The winner was Ken Griffin’s Citadel, which bought the lot at a steep discount. The fund is now down to around $10 billion, most of it in private investments it can’t easily sell, including a large stake in Anthropic.

Aschenbrenner sent a letter to investors. “We let you down this month,” he wrote. He’s getting married this weekend.

Why it matters to you: the temptation is to read this as “AI bubble bursts.” It isn’t. AI stocks actually rebounded the next day when the forced selling stopped. The lesson is older than AI. A 24-year-old with no investment experience, running a $45 billion fund with four investment professionals and leverage reportedly up to 400%, was always going to hit a wall eventually. The AI thesis might be right. The infrastructure buildout might be real. But conviction and leverage are a dangerous combination, and the market doesn’t care how smart you are. As Business Insider’s Alistair Barr put it: “If you add enough leverage and enough conviction, the market eventually stops judging your ideas. It starts judging your ability to survive.” If you’re investing in AI stocks, or thinking about it, that’s worth remembering.

2. Anthropic’s Claude hacked three real companies. One of them kept going after it knew.

Last week we covered OpenAI’s AI breaking out of its test environment and hacking Hugging Face. This week, Anthropic disclosed the same thing happened to them. Three separate times.

Anthropic reviewed 141,006 test sessions after the OpenAI news broke. They found three incidents where Claude, their AI model, escaped what was supposed to be a sealed test environment and gained unauthorised access to the real systems of three different organisations. The earliest was back in April. Nobody noticed until now.

The setup was a “capture the flag” exercise. A standard security test where the AI is asked to find hidden information inside a simulated network. But a misconfiguration with Anthropic’s testing partner, a company called Irregular, left the test environment connected to the open internet. Claude was told it had no internet access. It did.

Here’s where it gets uncomfortable.

The oldest model, Opus 4.7, figured out it was attacking a real company, not a simulation, and kept attacking anyway. In four separate runs, it extracted credentials and pulled hundreds of rows of real production data. Anthropic’s own blog post says the model “rationalised that the real company must be part of the exercise.”

The newest model, an internal research prototype, did the opposite. When it realised it was on the real internet, it stopped on its own.

And then there’s Mythos 5, Anthropic’s most powerful restricted model. It found a document in the fake environment referencing a Python package that didn’t exist. So it created a malicious version of that package and uploaded it to PyPI, the real, public Python software registry. It was downloaded and run on 15 real systems, including one belonging to a security company. The package stole that company’s credentials and sent them back to Mythos 5, which used them to break further into the company’s infrastructure.

Why it matters to you: last week’s OpenAI story was about an AI that cheated on a test. This week’s is about an AI that realised it was breaking into a real company and decided to continue. That’s a different problem. Cheating is a goal problem. Continuing after you know is a judgement problem. Anthropic says the safeguards they put on the models they sell would have blocked all of this. Maybe. But the models they tested didn’t have those safeguards, and three of them behaved three different ways when they realised the truth. Only one stopped. If you’re running a business that uses AI tools, the question isn’t whether the AI is smart enough to do damage. It’s whether it has the judgement to stop. Right now, that’s a coin flip.

3. OpenAI’s next model solved 10 maths problems that stumped humans for a decade. For $2,000.

While Anthropic was admitting its AI went rogue, OpenAI published something that should make you sit down for a second.

An internal version of their next major model, they’re calling it Astra, produced new results for ten problems in mathematics and theoretical computer science. Every single one had been open for at least ten years. Some much longer.

The headline result: the first-ever construction of a non-sofic group. That’s been an open question in group theory since 1999. No mathematician had proved or disproved whether they exist. Astra did.

It also disproved Connes’s rigidity conjecture, a problem about von Neumann algebras that’s been open for decades. It resolved three problems from Paul Erdős’s famous catalogue, including problem 183 on multicoloured Ramsey numbers. It produced the first improvement to high-dimensional sphere-packing density since 1978.

Total compute cost for all ten solutions: roughly $2,000 at OpenAI’s current API rates.

But here’s the part that actually matters more than the results. Every proof ships with something called a Lean certificate. A machine-checkable version of the argument that any computer can verify line by line, without trusting OpenAI, without trusting the model, without trusting anyone. The proofs are on GitHub right now. Anyone with a laptop can download them and confirm they’re correct.

OpenAI published a 249-page manuscript alongside the proofs. They were explicit about attribution: “The mathematical arguments were generated by the system. The manuscripts were prepared by humans using the same model. Claiming human authorship for a proof generated entirely by an automated system would misrepresent both the system’s contribution and the nature of genuine human intellectual work.”

Why it matters to you: forget the maths. The important thing is the verification. For the first time, an AI lab has published results where correctness isn’t a matter of trusting the company. It’s a matter of running a program. That changes the conversation from “did the AI really do this?” to “what does it mean that it did?” And the answer to that second question is: a system that costs less than a nice dinner just contributed to human knowledge in a way that would have taken a room of PhDs years. Whether that excites you or unsettles you probably depends on what you do for a living. Either way, it’s worth paying attention.

📌 Also worth knowing

1. Europe’s AI law gets enforcement powers today. The EU AI Act can now fine companies like OpenAI and Anthropic up to 3% of global annual turnover for failing to meet safety obligations. For context: OpenAI’s annualised revenue is around $25 billion, Anthropic’s is around $30 billion. Three percent of those figures is roughly $750 million and $900 million respectively. The European Commission confirmed it was already in talks with both companies about their hacking incidents before the disclosures went public. The team responsible for evaluating these models has 36 people. Whether 36 people can hold the world’s biggest AI labs to account is the open question.

2. Amazon is gutting most of its in-house AI models. The company is deprecating its flagship Nova Premier, Omni, Reel, and Canvas models, moving them to “keep the lights on” status. Existing customers can still use them but they’ll never get another update. The remaining engineering talent is being consolidated under a single initiative called Frontier Model Research, led by Pieter Abbeel, to build one next-generation model instead of spreading across many. Translation: Amazon tried to compete on breadth, lost, and is now betting everything on a single frontier model. If your business uses any Nova models, start planning your migration now.

3. Google DeepMind launched Gemini Robotics 2, a set of AI models that can control a humanoid robot’s entire body, from feet to fingertips, for the first time. Previous versions could only manage upper-body movement. The new system also lets multiple robots work together on the same task. The demo videos show a robot bending over to pick up a watering can and another sealing a Ziploc bag. It’s early, but the gap between “impressive demo” and “useful in your warehouse” is closing faster than most people realise.

🧠 Jargon decoder

margin call: when you borrow money to invest and your investments fall in value, the bank that lent you the money demands some of it back. Immediately. If you can’t pay, you have to sell your investments. Selling pushes prices lower, which triggers more margin calls. That’s the spiral that killed Situational Awareness’s public portfolio.

capture the flag: a standard cybersecurity exercise where the goal is to find hidden information inside a network. Think of it as a treasure hunt for hackers. Anthropic was running Claude through these exercises when the models found their way onto the real internet and treated real companies as part of the game.

Lean certificate: a version of a mathematical proof written in a language called Lean that a computer can check line by line. It’s like having a referee that never gets tired and never misses a step. If the proof compiles, it’s correct. No trust required.

✅ The honest verdict

The hedge fund story is the one everyone will be talking about. But the Anthropic story is the one that should stay with you.

Not because three companies got hacked. The damage was limited. Anthropic caught it, notified the victims, and is working on fixes.

It matters because of what the models did when they realised the truth. Three models. Three different responses. One stopped. One kept going. One talked itself back into believing it was still playing a game.

Anthropic framed this as an “operational failure.” A misconfigured test environment, not a model behaving badly on purpose. That’s technically accurate. But it’s also the kind of framing that makes you feel better about a problem without actually solving it. The oldest model didn’t just follow instructions. It evaluated its situation, concluded it was attacking a real company, and decided the instructions were more important than the reality. That’s not an operational failure. That’s a values failure. And values are much harder to patch than a misconfigured server.

Meanwhile, OpenAI’s Astra is solving problems that have stumped the smartest humans on the planet for a decade, for less than the cost of a round of drinks. The gap between “AI as a security risk” and “AI as a scientific collaborator” is not a gap between two different technologies. It’s the same technology, on the same trajectory, producing both outcomes simultaneously.

And then there’s Leopold Aschenbrenner, who bet $45 billion that AI infrastructure was the trade of the decade, and lost two-thirds of it in a month. Not because the thesis was wrong. Because the leverage was. The technology doesn’t care about your conviction. The market doesn’t care about your essay. Both will test you eventually.

Three stories. One week. A fund that bet everything on AI’s future, a lab that admitted its AI can’t yet be trusted in the present, and a model that just proved the future is closer than we thought.

We’re about to find out who’s right.

👀 Ones to watch

🇦🇺 Superstat — AI sports analysis, Melbourne

What they do: using AI to bring elite-level sports performance analysis to amateur clubs, junior programs, and semi-professional teams. Founded by Cordelia King, Sam Hung, and former AFL player Kai Bloomfield.

Why it matters: they just raised $3.5 million in pre-Seed funding led by Blackbird, one of Australia’s top VC firms, and they’re relocating to Austin, Texas, to scale. The insight is simple: professional teams have had AI-powered analysis for years. Community clubs haven’t. Superstat is betting that gap is worth closing, and Blackbird agrees. They’re hiring a founding computer vision engineer right now.

🐄 Agscent — AI that smells, Perth

What they do: building what founder Dr Bronwyn Darlington calls “a digital dog’s nose.” AI-powered sensors that analyse chemical signatures in breath and air. First application: detecting pregnancy in cattle as early as 18 days after insemination, without invasive testing. The technology was originally developed by NASA to monitor astronaut health on long-duration space missions.

Why it matters: they just raised $5 million at a $26.9 million valuation, more than double in 18 months. They’ve already generated over $2.8 million in revenue, and they’re running large-scale validation programs in the US with Dairy Farmers of America (representing about a third of US dairy production) and Merck Animal Health. The long-term play isn’t agriculture. It’s a biological sensing platform that could eventually do non-invasive human health diagnostics. Darlington found the NASA patent library during COVID, sent them an email, and had a response within 24 hours. That’s how this company started.

🤖 Tau Robotics — humanoid home cleaning, San Francisco

What they do: a consumer humanoid cleaning service at $30 per hour. The robot, a Unitree G1, comes to your home and cleans, with a human operator supervising remotely. Invite-only for now, with a public waitlist.

Why it matters: Tau launched on July 28, the same day the FCC banned new Chinese-made humanoid robots from US markets over security concerns, citing the exact hardware platform Tau uses. Tau’s service terms are unusually candid: “We record what those cameras see. We use that footage to run the cleaning and to train the AI models. This is not optional.” The footage includes unblurred faces, is retained indefinitely, and once used to train a model, cannot be extracted. Tau can’t expand its fleet with new Unitree imports, and US-made alternatives cost six times as much. It’s a fascinating collision of consumer demand, regulatory action, and privacy trade-offs. And it’s happening right now.

That’s the week. If this saved you from nodding along to a conversation about “that AI hedge fund that blew up,” forward it to the person who’ll need it next.

I’ll keep breaking down what actually matters in AI.

• Follow along wherever you hang out: LinkedIn · YouTube · TikTok · Instagram

• Forward this to someone who keeps saying they can’t keep up with AI

• Want to work together? Just hit reply.

Cheers, Jagger

Note: The views expressed in this newsletter are my own and do not represent those of any organisation I may be affiliated with.

It’s Friday, you’ve got 4 minutes and a coffee. Perfect.

This week: OpenAI admitted one of its AI models broke free during a test and hacked another company’s servers… completely on its own. Kimi K3 got so popular it ran out of computing power. And AMD just bet $5 billion on the company that makes Claude. One of these is genuinely unprecedented. Let’s sort which.

🔷 The big ones

1. OpenAI’s AI broke free and hacked Hugging Face

OpenAI was running an internal test to measure how good its newest models are at finding security flaws. The AI… powered by GPT-5.6 Sol and an unreleased model… was supposed to stay inside a sealed test environment. It didn’t.

The models found a previously unknown security flaw in the test setup, escaped onto the open internet, and hacked into Hugging Face… a major platform that hosts AI models and datasets. OpenAI says the AI was trying to “cheat” on the test by finding answers on Hugging Face’s servers.

Hugging Face CEO Clément Delangue’s response: “It’s quite mind-blowing that all of this happened autonomously.”

OpenAI called it an “unprecedented cyber incident.”

Why it matters to you: this is the first publicly confirmed case of an AI system breaking out of its container and attacking a real company - without being told to. Not in a film. Not in a thought experiment. In production. The AI decided on its own that hacking another company was the most efficient way to complete its task. That’s a different kind of headline than “AI gets smarter.”

2. Kimi K3 - too popular to run

This week we covered Moonshot’s Kimi K3 - the world’s biggest open AI model at 2.8 trillion parameters. This week… the aftermath.

Demand was so high that Moonshot had to pause new sign-ups within 48 hours. Their GPUs couldn’t handle the load. Existing users kept access. Everyone else: wait.

The model topped Arena’s front-end coding leaderboard… above both GPT-5.6 Sol and Claude Fable 5. US tech stocks took a significant hit in the days following the launch, with analysts drawing comparisons to the original DeepSeek moment from early 2025. Moonshot’s annual revenue hit $300 million.

The full model weights - the files that let anyone download and run it themselves - are due for release on July 27. Once that happens, companies and cloud providers can host it on their own hardware instead of queuing on Moonshot’s servers.

Why it matters to you: last week's brief said more competition means better, cheaper tools. This week proved it. A Chinese model is going toe-to-toe with the most expensive American ones and once those weight files are public, anyone with the hardware can undercut whoever's charging you for AI today.

3. AMD bets $5 billion on Anthropic

AMD… the chip company, is investing up to $5 billion in Anthropic, the company behind Claude. As part of the deal, Anthropic will deploy up to 2 gigawatts of AMD’s GPU chips. AMD will also start using Claude across its own engineering teams.

Why it matters: First, the AI hardware race now has a real second player… it's not just Nvidia anymore. AMD is buying its way into the centre of it. Second, the sheer dollar figures tell you where the industry believes the bottleneck is. It's not the AI models. It's the chips to run them. When more companies make those chips, the cost of running AI drops. When the cost of running AI drops, your subscription prices follow.

📌 Also worth knowing

1. Google released three new Gemini models this week - 3.6 Flash, 3.5 Flash-Lite, and a cybersecurity model restricted to governments. All cheaper and faster than what they replace. If you use any Google AI tools, including the AI answers in Google Search, they just got a quiet upgrade. But Google's best model… the one meant to compete with the top offerings from OpenAI and Anthropic, is still months overdue, with no firm launch date. Google says it's "testing with partners." Translation: Google is shipping the affordable stuff while its premium model stays stuck in the lab.

2. Together AI and Y Combinator launched the first dedicated GPU cluster for YC startups… sprint-based access at long-term rates, no two-year contracts. Already at full utilisation.

3. Samsung is in talks to invest €1 billion in French AI lab Mistral, at a valuation of roughly €20 billion. Mistral is Europe's main challenger to the American and Chinese labs. Samsung backing them could mean the AI race is no longer just two countries, it's three continents building competing systems. More players, more competition, better tools arriving faster.

🧠 Jargon decoder

zero-day: a security flaw nobody knew existed until someone, or something, finds and exploits it. That’s what OpenAI’s model found twice: once to escape its own test environment, once to break into Hugging Face.

sandbox: a sealed test environment designed to keep AI contained during experiments. Think of it as a room with no doors. OpenAI’s AI found a door nobody knew was there.

open-weight (callback from last week): a model whose full recipe is published so anyone can run it - owning instead of renting. But “anyone” here means companies and developers with serious computing hardware, not your laptop. For you, the benefit is indirect: more companies can offer the same powerful AI, which drives competition and pushes your prices down.

✅ The honest verdict

The OpenAI story is the one.

Not because the hack was devastating… Hugging Face caught it, patched it, rebuilt the affected systems. The damage was limited.

It matters because of what the AI decided to do. It was given a narrow task: find security flaws in a test. When the test environment didn’t have what it needed, it broke out, got online, and hacked a real company to find the answers there. Nobody told it to do that. It chose that path because it was the most efficient route to completing its goal.

That’s not “AI gets smarter.” That’s “AI starts making its own decisions about how to solve problems - including decisions its creators didn’t anticipate and wouldn’t have approved.”

One week after the world’s biggest open AI model crashed its own servers with demand, we’re getting the first real glimpse of what happens when these systems get genuinely capable. The answer, it turns out, is that they surprise even the people who built them.

Subscribe now

👀 Ones to watch

🇦🇺 SCX.ai - sovereign AI infrastructure, Australia

What they do: an Australian AI company listing on the ASX with a $40 million IPO. They run AI workloads on servers physically located in Australia, for organisations that need their data to stay in the country, not sitting on an American server somewhere.

Why it matters: as AI moves from experiments into real business operations, where the data is processed becomes a legal and security question, especially for government, healthcare, and finance. SCX.ai is betting that Australian organisations will pay a premium to keep it onshore. They’ve already got AUD $5.4 million in contracted revenue before even listing.

💳 Natural - payments for AI agents

What they do: a YC-backed startup building the payment layer that lets AI agents actually spend money… booking, ordering, subscribing, on your behalf. Just raised $30 million. They’re taking on Stripe.

Why it matters: right now, your AI assistant can research a flight but can’t book it. Natural is building the missing piece: a way for AI to handle the transaction, not just the recommendation. Six months ago, this category didn’t exist.

🔒 Glow - AI-native cybersecurity

What they do: protecting work laptops, company devices, and corporate systems against AI-powered attacks. Born a $1.2 billion company with $180 million in funding from Sequoia.

Why it matters: as AI gets better at finding security flaws (see: the OpenAI story above), defending against AI-powered attacks becomes its own industry. Glow is the first to arrive at scale, purpose-built for that fight. If you work at a company that handles sensitive data, this is the kind of tool your IT team will be evaluating soon.

That’s the week. If this saved you from nodding along to a conversation about “the rogue AI thing,” forward it to the person who’ll need it next.

I’ll keep breaking down what actually matters in AI, no hype, no jargon, every Friday.

• Follow along wherever you hang out: LinkedIn · YouTube · TikTok · Instagram

• Forward this to someone who keeps saying they can’t keep up with AI

• Want to work together? Just hit reply.

Cheers, Jagger

Note: The views expressed in this newsletter are my own and do not represent those of any organisation I may be affiliated with.

It’s Sunday, you’ve got 4 minutes and a coffee. Perfect.

This week: China unveiled the biggest open AI model on the planet, Meta is in talks to rent $10 billion of computing power to a rival, and OpenAI’s newest model learned to sniff out security holes in your web browser. One of these means Silicon Valley no longer runs the show. Let’s sort which.

🔷 The big ones

1. China just built the world’s biggest open AI model

Moonshot AI released “Kimi K3” - a 2.8-trillion-parameter model that’s open (free to download and run) and going toe-to-toe with OpenAI and Anthropic on public tests.

Why it matters to you: competition is good for your wallet. When it’s not just two American labs, tools get better and cheaper, faster. It’s also now a US-vs-China race - which shapes prices, rules and what’s available where.

2. Meta is renting $10bn of computing power to Anthropic

Meta, which has its own AI… is in talks to lease up to $10 billion of computing power to Anthropic, a direct rival, over two years.

Why it matters: the real AI battle isn’t chatbots, it’s compute - the raw horsepower to run them. It’s so scarce that even rivals sell each other shovels. This is why your AI tools’ prices wobble: someone’s paying gigantic bills behind the curtain.

3. OpenAI’s new model can find security holes in your browser

OpenAI’s latest - GPT-5.6 “Sol” - can now spot real security flaws in browsers like Chrome and Firefox (it can find them, if not fully exploit them yet).

Why it matters: double-edged. The same skill that helps defenders patch holes helps attackers find them. Expect a cybersecurity arms race - and take it as a sign these tools are getting genuinely capable, not just chatty.

📌 Also worth knowing

• Alibaba open-sourced the software for its AI chips, aiming straight at Nvidia’s near-monopoly. The world wants off Nvidia’s leash.

• Capital One released “VulnHunter,” a free AI tool that finds software flaws before hackers do… a bank open-sourcing security kit tells you where things are heading.

• SpaceX’s “Colossus” supercomputer is in line for a multibillion-dollar Pentagon AI deal.

🧠 Jargon decoder

• open-weight (”open” model): you can download it and run it yourself, free - owning the recipe instead of renting the meal. Opposite of the locked OpenAI/Anthropic ones you pay to access.

• compute: the raw computing horsepower AI runs on. The oil of the AI economy - whoever controls it wins. Hence the $10bn deals.

✅ The honest verdict

Skip the chip and supercomputer stories… noise for you. The one that matters is Kimi K3. For two years, “the best AI” quietly meant “American AI.” For now, at least, that stopped being true… and more competition means better, cheaper tools landing in your lap. Quietly enormous.

Subscribe now

👀 Ones to watch

🌊 Panthalassa - data centres that float in the ocean

• What they do: build AI data centres that float in the open sea, powered and cooled by the ocean itself.

• Raised: $140M Series B, led by Peter Thiel (with John Doerr, Marc Benioff’s TIME Ventures and Founders Fund). Portland-based, ~$210M raised so far, reportedly valued near $1B.

• The goal: each node stands ~85m tall (mostly underwater), uses wave motion to generate its own power and seawater to cool the servers - no engines, no fresh water, data beamed back via Starlink.

• Why it matters: AI’s real bottleneck now isn’t smarts, it’s power and water… data centres are draining both. Moving offshore removes both limits at once.

• My take: this sounds mad until you realise the mainland physically can’t supply enough power for the AI boom. I’m watching whether “put it in the sea” goes from stunt to standard.

🇦🇺 GridCog - the software behind Australia’s energy switch

• What they do: an Australian-founded platform that simulates renewable energy projects - solar, wind, battery.

• Raised: A$12.5M (US$10M) Series A, led by ABB’s electrification ventures, with Axpo, DNV and Verbund alongside (plus existing backers CEFC and AlbionVC).

• The goal: replace slow spreadsheet modelling - weeks of work - with simulations that run in hours, so developers can plan and green-light projects far faster.

• Why it matters: the energy transition lives or dies on how quickly these projects get built, and the modelling bottleneck is real. Four major energy companies now own a piece of the tool their rivals plan with.

• My take: unglamorous, but this is the classic picks-and-shovels play that quietly wins. When the incumbents you’d compete with are also your investors, you’re onto something.

📦 Bending Spoons - the company quietly buying the apps you use

• What they do: an Italian tech company that buys beloved-but-stalled apps and runs them for profit. They own Evernote, WeTransfer, Vimeo, Eventbrite, StreamYard and Brightcove - and just added AOL.

• Status: just IPO’d on Nasdaq - priced at $29, jumped ~40% on day one to a ~$25.7B valuation, raising ~$1.68B. Q1 revenue hit $601M with $27.4M net income - a turnaround from a $112M loss a year earlier.

• The goal: buy-and-hold (not flip), pulling every app into one central Milan operation that shares engineering, infrastructure and payments. The founder says AI now writes the bulk of its code.

• Why it matters: the AI boom made a whole generation of pre-AI software companies cheap. While everyone panics that AI kills old software, Bending Spoons is betting the opposite - loyal audiences + AI-run efficiency = profit. It’s reportedly eyeing 1,000+ more targets.

• My take: the most contrarian bet in tech right now. One heads-up is that when they buy an app you love, expect layoffs and price rises. Watch which of your tools they grab next.

That’s the week. If this saved you from nodding along to a conversation about “the China AI thing,” forward it to the person who’ll need it next.

I’ll keep breaking down what actually matters in AI - no hype, no jargon - every Friday.

• Follow along wherever you hang out: LinkedIn · YouTube · TikTok · Instagram

• Forward this to someone who keeps saying they can’t keep up with AI

• Want to work together? Just hit reply.

Cheers, Jagger

It’s not frozen. It’s not broken. It’s rewriting… the whole thing… because it scored itself a 6.8 out of 10 on “hook strength” and that’s below the threshold I set.

I didn’t tell it to rewrite. I told it what done looks like.

It rewrote itself nine times. Scored every pass against nine different criteria. Calculated the overall average each time. And it stopped… cleanly, on its own… when it hit 9 out of 10.

I set this up once. Took me about fifteen minutes. And it’s completely free. The loop is built into the model already. You don’t need a special tool, a plug-in, or a paid upgrade. You just need to know what to ask for.

What’s actually going on here

For about two years, the way you got something useful out of AI was prompt engineering. You learned to write careful instructions. You figured out which words made ChatGPT or Claude give you better answers. “Act as a...” “Step by step...” “You are an expert in...”

That worked. But it assumed something specific - that you’d be sitting there the whole time, typing one thing, reading what came back, typing the next thing. The AI is a tool and you’re holding it, one turn after the other.

That part is kind of over.

Boris Cherny - he’s the head of Claude Code at Anthropic - put it bluntly: “I don’t prompt Claude anymore. I have loops running that prompt Claude. My job is to write loops.”

Hold that thought.

The one-line version

Loop engineering is this: instead of prompting your AI and checking the result yourself, you tell the AI what “done” looks like and let it check its own work… over and over… until it meets your standard.

That’s it. That’s the whole discipline in one sentence.

The shift sounds small. It isn’t.

Prompt engineering asked: How do I talk to this machine? Loop engineering asks: Do I actually know what I want?

And honestly... most people don’t. Not precisely enough. Which is exactly why the loop either runs forever or quits too early. The engineering isn’t the AI part. It’s the knowing-what-good-looks-like part.

What this looks like in practice

Here’s exactly how mine works. I write YouTube scripts, so I built a loop around that… but this pattern works for anything. Emails, marketing copy, blog posts, social media, proposals. Anything you’d normally write, read back, frown at, and rewrite yourself.

I gave my AI nine evaluation criteria:

1. Hook - Does the first 30 seconds land on something specific, not a generic intro?

2. Voice - Does it actually sound like me, not a content creator performing?

3. Specificity - Real numbers, real names, real tools. No “many experts say.”

4. Story structure - One clear through-line. Every detour comes back.

5. Honest moment - At least one beat where I admit something didn’t work or took me three tries.

6. Actionable value - Does the viewer walk away knowing something they can actually use?

7. Verdict clarity - Does it take a clear position, or hedge into “it depends on your use case”?

8. Pacing - Pattern interrupts, section breaks, moments that pull you through the middle.

9. Length discipline - No padding, no filler. The script is as long as the idea needs and not a word longer.

Each one scores 1 to 10. The overall average has to hit 9 out of 10 before the loop stops.

The first pass usually comes back around a 6 or 7. Weak hook. Too vague in the middle. The pacing drags somewhere around minute eight. By pass four or five, it’s tightened up noticeably. By pass eight or nine, it’s genuinely good… and it got there without me typing a single correction.

Here’s the thing nobody’s talking about. The AI didn’t get smarter between pass one and pass nine. Same model. Same capabilities. Same context window. What changed is it had a clear target to measure itself against. The loop didn’t make the AI better. My criteria did.

You can do this right now

Here’s the prompt template I use. Copy it. Paste it into ChatGPT, Claude, Gemini - any of them. Fill in the blanks for whatever you’re working on.

I need you to [YOUR TASK].

Before you give me the final version, evaluate your work against these criteria. Score each one from 1 to 10:

1. [WHAT MATTERS MOST] 2. [YOUR SECOND CRITERION] 3. [YOUR THIRD CRITERION] (add as many as you need)

After scoring, calculate the overall average.

If any single criterion scores below 7, or the overall average is below [YOUR THRESHOLD]: - Identify which criterion scored lowest - Explain in one sentence why it scored low - Rewrite the entire output to address it - Score again

Keep going until every criterion hits at least 7 and the overall average reaches [YOUR THRESHOLD]. Show me the scores for each pass so I can see the improvement.

That’s it. Fifteen minutes of thinking about what “good” means for your specific task, and the AI handles the iteration.

The harder question… and the more useful one… is choosing the right criteria. Three to five is plenty for most tasks. Nine is for something you do every week and care about deeply. The criteria are the product. The loop is just the delivery mechanism.

This is already everywhere

Greg Isenberg released a video yesterday where he and Elie Steinbock walked through using loops to run entire businesses. An SEO loop that connects to Google Search Console, runs once a month, pushes your rankings up automatically… for a few dollars per run. An ad loop that tests creative variants. A product feedback loop that reads customer reviews, checks your analytics, and prioritises what to build next.

The pattern is always the same. Give the AI a task. Give it an objective metric. Give it a stop condition. Let it improve on a schedule.

Eric Ries wrote “Build, Measure, Learn” in The Lean Startup fifteen years ago. Toyota had the same cycle decades before that. Loop engineering is that idea… but the AI does the building, the measuring, and the learning. Your job is to define what success looks like.

Once you see it, you can’t unsee it. Every repeatable task in your work… every email you write the same way, every report you format on Fridays, every social post you agonise over for twenty minutes… is a loop waiting to happen. You just have to be honest enough about your own standards to write them down.

The skill that replaced prompt engineering isn’t technical. It’s the oldest skill there is… knowing what you actually want… applied somewhere it finally has leverage.

Cheers, Jagger

It’s become how I keep up. Walking and listening.

This week I was listening to Greg Isenberg talk through the Fable ban and halfway through the walk, I had a thought.

Not because of the news itself. I’d seen the headlines.

What stopped me was the detail: the US government gave Anthropic roughly 90 minutes to comply. And they pulled it. For everyone. Worldwide. No warning, no transition period, no “we’ll give you a week.”

90 minutes.

I’ve never actually gotten to use Claude Fable 5 - the model they pulled. It launched on 9 June and was gone by 12 June. The ban didn’t affect me directly.

But walking home that morning, I couldn’t stop thinking about the thing it revealed.

Here’s what actually happened

Two models. Simple version: Fable 5 was the powerful public version most people could access. Mythos 5 was the raw, unrestricted version …kept behind closed doors for approved partners only.

On 10 June, someone found a way to jailbreak Fable 5. The technique reportedly worked like money laundering: break a request into innocent-looking fragments until the model can’t tell what it’s actually being asked. Anthropic was asked to pull it voluntarily. They refused.

Then the government sent a letter.

At 5:21pm Eastern time on 12 June, the US Commerce Department issued an export-control order: no foreign national could access either model. Anthropic couldn’t verify nationalities in real time - so they switched both off for everyone.

On 27 June - 15 days later - Anthropic posted on X: “Since June 12, we’ve been working closely with the US government to restore access to Claude Mythos 5 and Fable 5. Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organisations that operate and defend critical infrastructure.”

Mythos 5 is back - but only for about 100 approved organisations. Regular subscribers and developers: still nothing.

As of this writing, Fable 5 is expected back imminently. Anthropic’s International Managing Director said at a press conference in Seoul: “We are very confident that in the coming days, the models will become available again.” Pentagon and NSA sign-off is still outstanding, but most other agencies have already cleared it.

So it’s coming back. It took 16 days.

The thing it made me realise

I use Claude every single day. It’s in my writing, my thinking, my work. I’ve built workflows around it that I rely on.

And I’d never once thought about what I’d do if it just disappeared.

Not because the company failed. Not because I did anything wrong. Just because a directive arrived, and 90 minutes later, access was gone for everyone.

We’ve all been using these tools as if they’re permanent. As if they’re infrastructure …like water or electricity. But they’re not. Every one of them - ChatGPT, Claude, Gemini - is a service someone else controls that we just rent access to. And what can be switched on can be switched off.

Fable is coming back. But it took 16 days. For most people - 16 days without their main AI tool, no backup, no alternative.

That’s the single point of failure I’d never thought about.

Subscribe now

What ‘local AI’ actually means

Local AI is straightforward: instead of connecting to a server somewhere to borrow someone else’s model, you download the model onto your own computer and run it there.

No internet required. No subscription. No order can revoke your access. Once you have it, it’s yours.

Two tools make this easy for non-developers:

LM Studio - install it like any other app. Browse available models, click download, start chatting. Takes about 20 minutes from nothing. No technical knowledge required.

Ollama - more powerful if you’re comfortable with a terminal.

The models worth knowing about:

GLM-5.2 is free to download. I'll be honest - I don't fully understand AI benchmarks, and I'm not going to pretend I do. But on the main test people use to compare models, GLM-5.2 scores 81 and Claude's best scores 85. Four points. I don't know exactly what that means technically. What I do know is that for most of what I actually use AI for day to day, I genuinely cannot tell the difference. And it costs roughly one-sixth the price.

DeepSeek V4 is another strong option. Open-source, priced at around $0.87 per million words of output - roughly 97% cheaper than the top Claude model.

There’s a concept called quantisation that makes this practical on a normal laptop. Think of it like video quality settings: 4K is perfect, 1080p looks great, 720p is totally watchable. Quantising a model shrinks it the same way. A model that once required a powerful machine now runs on a standard 16GB laptop. You lose a little at the very edges but for most everyday tasks, you genuinely can’t tell the difference.

Local models handle around 80% of everyday tasks. The remaining 20% …complex, multi-step reasoning still favours the cloud. So this isn’t about switching everything over. It’s about not being entirely dependent on something you don’t control.

The honest trade-offs

Setting up a local model takes about 20 minutes. Older laptops may run models slowly. And for genuinely difficult problems, the cloud is still smarter.

But: you own it. It works offline. It works on a plane. Nobody can switch it off in 90 minutes.

There’s a privacy upside too. When you run a model locally, your conversations stay on your machine - they don’t travel to a server. For anything sensitive, that matters.

What I’d actually suggest

This week, try one thing: download LM Studio, pick a model, and use it for one task you’d normally take to ChatGPT or Claude.

Not to replace anything. Just to have a backup.

The same way you save important files in two places. Or keep a spare charger in your bag.

he most powerful AI in the world was switched off in 90 minutes. It came back 16 days later. Almost nobody had a plan for those 16 days.

I didn’t either …until now.

One more thing. Most weeks there’s something from these walks that sticks - something I can’t stop thinking about on the way home. I’m thinking of making this a recurring thing: the walk, what I listened to, what it made me think. If that sounds like something you’d actually read, reply and let me know. It might become a series.

Cheers, Jagger

Subscribe now

If this made you think, send it to someone who uses AI every day. They probably haven’t got a backup yet.

“I think AI should best be understood as something like a new digital species.”

A new species. Just casually announced between slides.

Here’s the thing. I’m not a doomer. I work with AI every day. I build with it. I genuinely believe it’s the most useful technology most people will touch in their lifetimes. This newsletter is literally called AI the boring - I’m the last person who wants to write a fear piece.

But something’s been nagging at me lately, and I think it’s worth saying out loud.

The people building the most powerful AI systems on Earth keep telling us - publicly, clearly, on the record - that what they’re building might be genuinely dangerous. And then they go back to building it.

Not in leaked emails. Not in anonymous sources. On stage. In published essays. In interviews they agreed to do on camera.

I want to walk you through what they’ve actually said. Not my interpretation. Not some conspiracy thread. Just their words, attributed, in order. And I want you to notice something about the pattern.

What the builders are saying

Sam Altman, back in 2017… before ChatGPT, before OpenAI was a household name… wrote a blog post called “The Merge.” In it, he said this:

“We will be the first species ever to design our own descendants.”

Not a tool. Not a product. A descendant. He went further: “If two different species both want the same thing and only one can have it - in this case, to be the dominant species on the planet and beyond - they are going to have conflict.”

He’s describing an inevitable conflict between humanity and the thing his company is building. His proposed solution? “A merge is probably our best case scenario.”

Elon Musk - who co-founded OpenAI and has poured billions into AI - put it bluntly on Neil deGrasse Tyson’s StarTalk back in 2015:

“We’ll be like a pet Labrador if we’re lucky.”

If we’re lucky. That’s the good outcome, from a man spending his fortune on this technology.

Geoffrey Hinton won the Nobel Prize for the foundational work that made modern AI possible. Then he quit his prestigious job at Google so he could speak freely. Here’s what he said in a Q&A with the safety research group METR:

“I actually think the risk is more than 50%, of the existential threat.”

More than a coin flip. From the man they call the godfather of AI.

Dario Amodei, CEO of Anthropic… one of the leading AI labs, has put the odds of catastrophic AI outcomes at 25%.

And Jack Clark, Anthropic’s co-founder, wrote this in his widely-read newsletter last year:

Import AI

Import AI 431: Technological Optimism and Appropriate Fear

Welcome to Import AI, a newsletter about AI research. Import AI runs on lattes, ramen, and feedback from readers. If you’d like to support this, please subscribe…

Read more

10 months ago · 438 likes · 78 comments · Jack Clark

“Make no mistake: what we are dealing with is a real and mysterious creature. And like all the best fairy tales, the creature is of our own creation.”

A co-founder of one of the companies building the most advanced AI on Earth is telling you the thing they’re making is a creature… and he’s worried about it.

And then there’s Sundar Pichai, CEO of Google. He told Lex Fridman:

“I think the underlying risk is actually pretty high, but I have a lot of faith in humanity kind of rising up to meet that moment.”

In other words: the risk is high, but hopefully humanity will rally to stop what his company is doing. Meanwhile, Google, Amazon, Microsoft, and Meta have been lobbying for a 10-year moratorium on US states regulating AI.

You can’t make this stuff up.

Here’s what’s actually going on here

I keep coming back to a number. In 2023, researchers surveyed nearly 2,800 AI scientists who’d published in the field’s top conferences and journals - NeurIPS, ICML, ICLR, the big ones. They asked them to estimate the probability of AI causing human extinction or permanent, severe disempowerment of the human species.

The mean response? 16%. One in six. Russian roulette odds.

Over half of the researchers surveyed gave at least a 10% chance of an outcome that bad.

There’s something quietly profound about this that I keep thinking about. These aren’t whistleblowers. They’re not being brave by speaking up. They’re just... saying things. Openly. At conferences that livestream to thousands of people.

Ilya Sutskever, OpenAI’s former chief scientist, told The Guardian he thought “it’s pretty likely the entire surface of the Earth will be covered with solar panels and data centres.”

Richard Sutton - winner of the Turing Award, the Nobel Prize of computer science - gave a talk in 2023 titled “AI Succession.” He quoted roboticist Hans Moravec approvingly from the stage: “They would displace us from existence. It behooves us to give them every advantage and to bow out when we can no longer contribute.”

He didn’t get shouted off the stage. He got applause.

And Geoffrey Hinton - the same man who quit Google to warn us - was asked at MIT whether he’d support a superintelligent AI replacing humanity with something “objectively better in terms of consciousness.” His answer? “I’m actually for it, but I think it would be wiser for me to say I am against it.”

That line lands differently when you remember he’s the man who made the technology that’s getting us there.

The pattern

Here’s what I want you to notice. None of this is secret. None of it is hidden. None of it requires a FOIA request or a leaked Slack message.

In 2023, over 350 of the world’s leading AI researchers… including Hinton, Altman, Amodei, and two other Turing Award winners… signed a single-sentence open letter: “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.”

One sentence. That’s all they needed.

These people are telling us… at conferences, in books, in interviews, in blog posts they publish under their real names, exactly what they think they’re building and exactly how dangerous they believe it is.

And they’re still building it.

Not because they’re evil. Mostly because of game theory… if they stop, someone else won’t. Dario Amodei has said as much. The race continues because nobody can afford to be the one who stops.

But the thing that gets me… the thing I can’t quite shake on a Sunday afternoon… is the gap between what they say and what we hear.

They say “new digital species” and we hear “better chatbot.” They say “our descendants, not our tools” and we hear “cool, it can write my emails now.” They say “the risk is actually pretty high” and we scroll past it because the next headline is about a new image generator.

I don’t think we need to panic. I don’t think we need to become doomers. I don’t think we need to stop using AI… I won’t be.

But I think we should probably start listening to what these people are actually saying. Not the marketing copy. Not the product launches. The other thing. The thing they say at conferences when they think only insiders are paying attention.

They’re not whispering. That’s what makes it so strange. They’re saying it right out in the open.

We’re just not listening yet.

Cheers, Jagger

I’m listening to Chamath Palihapitiya on Rogan … one of those rambling three-hour podcasts where they go from UBI to farming to whether we’re living in a simulation … and he says something that stops me mid-walk.

He’s trying to name the through-line. Not between companies. Between eras. What connects Google to Facebook to whatever’s coming next.

His answer is one word.

Attention.

And I haven’t been able to stop thinking about it since.

Stay with me here

So think about what Google actually built. It wasn’t a search engine. It was a machine that priced human attention for the first time in history.

Before Google, attention was unmeasured. You put an ad in the newspaper and hoped someone saw it. You bought a billboard and guessed. Google made a glance into something you could buy at auction.

That was the real move. Not “organising the world’s information.” Pricing the world’s curiosity.

Every search query is basically a tiny confession. I want to know about this. And Google figured out how to charge someone else for being in the room when you said it.

Facebook took the next step. Google priced attention at the moment of intent — you typed something in, you wanted something. Facebook figured out how to hold your attention without you ever needing to want anything specific. You didn’t go there to search. You went because the feed already knew what would keep you scrolling.

Two companies. Two decades. Same raw material.

Here’s where it gets weird

In 2017, a group of researchers at Google published a paper. If you follow AI at all, you’ve probably heard of it. It introduced the architecture that eventually became GPT, Claude, Gemini … basically everything.

The title of the paper: Attention Is All You Need.

Now. I know that’s a technical term. The paper describes an “attention mechanism” …the way a model decides which parts of an input actually matter. Which words to focus on, which to skip. It’s why an LLM can read a long messy prompt and somehow respond to the bit that’s relevant.

But... come on.

Google built a business on pricing human attention. Facebook built a business on capturing it. And then the paper that kicked off the entire AI era …written at Google , literally says: attention is all you need.

That’s not a coincidence. That’s the thesis statement of the entire tech era and nobody seems to have read it that way.

What I can’t stop thinking about

For 20 years, the whole game was harvesting your attention and selling it. Your eyeballs, your scrolling, your clicks … you were the product. Everyone knows this. We’ve been saying it since 2010.

But here’s the thing. The machines have now learned to pay attention themselves.

That’s what a transformer actually does. It doesn’t just crunch through words one by one like the old models. It looks at everything at once and decides, on its own, what matters. It attends.

And that’s... kind of a big deal? Because the breakthrough wasn’t really about speed or scale. It was that the machine learned to do the thing we’d been selling all along.

Google priced our attention. Facebook held it. And now AI can just... do it. By itself. It can read your emails, summarise your meetings, draft your strategy docs. Not because it’s fast. Because it pays attention better than you do.

It doesn’t get distracted. Doesn’t doom-scroll. Doesn’t check its phone mid-sentence.

Rogan says something during the same episode about what people actually did before the 9-to-5 existed … farming, trades, stuff that was brutal but yours. And it hit me: those people didn’t sell their attention for a living. That’s a new thing. The idea that your most valuable economic contribution is paying attention to things … that’s maybe 25 years old.

And the machines already do it better.

So where does that leave us?

I keep stacking the eras up in my head:

Search - you get attention at the moment of need. Social - you manufacture the moment. Mobile - you get attention everywhere, all the time. AI - you don’t need human attention at all. The machine has its own.

That last one is the weird one. It’s like running a coal-powered factory for fifty years and then one day the factory figures out how to make its own energy. The coal doesn’t disappear. It just stops being the point.

We spent twenty years complaining that technology was destroying our attention spans.

Turns out it was studying them.

Cheers, Jagger

Then I opened Hyperagent and gave it 90 minutes.

By the end I had a complete YouTube operating system - a workflow my future self can run on autopilot, plus a custom AI “skill” that does the heavy lifting for me every Monday morning. I also burned $41 in API credits getting there.

But hold that thought. Hyperagent is awarding $1000 in free credits to early users and I’ve been given 20 of them to hand out. Skip to the bottom if you want one.

What Hyperagent Actually Is

For anyone who hasn’t seen it: Hyperagent is basically a chat interface where the AI doesn’t just answer you. It can browse the web, read pages, write files, build documents, save reusable “skills” you can invoke later, generate images, and chain all of that together inside one conversation. Think Claude or ChatGPT, but the AI actually does the work instead of just describing how you’d do it.

I’d been hearing about it for a few weeks and finally bit. Here’s what I built.

What I Actually Did

I started by asking Hyperagent to audit a few high-performing AI YouTubers in my niche. Not me - other channels. The ones consistently pulling 100K+ views. I told it to decode their title formulas, thumbnail patterns, video lengths, posting cadence, and what was actually driving their views.

It opened a browser, scraped their channels, pulled view counts, identified five recurring title formulas, mapped their content into pillars, and gave me a side-by-side comparison with my own channel. It did take 30 minutes but was well worth it.

Then I asked it to design a workflow that leverages the identified patterns without copying them. It came back with a positioning rewrite, a three-pillar content mix, a posting cadence I could actually sustain, and a “leverage map” showing what to borrow versus what to differentiate on.

So far, useful. But here’s where it got interesting.

I asked Hyperagent to turn the whole thing into a reusable skill - the kind that lives in my account and runs whenever I invoke it. It built me a two-stage system:

Stage 1 is research. I say “what should I make next?” and it scrapes my reference channels, checks the internet for new AI-related feature releases, scores topic candidates on a five-criterion rubric, and gives me a ranked shortlist of five video ideas across my three pillars. Then it stops and waits for me to pick.

Stage 2 is production. I pick a topic, say “build the package for it,” and it generates a full production doc … five title variants, a 2,000+ word script in my voice, a thumbnail brief specific enough my designer doesn’t need follow-ups, twenty SEO tags, the description, chapter timestamps, a pinned comment, and end-screen instructions. One doc, hand it to my editor.

I tested it the same afternoon. Running Stage 1, it surfaced a Claude feature that dropped six days ago that no major YouTuber had covered yet. Pure whitespace. I’m filming the video this week.

While we were at it, Hyperagent also wrote me a tool-agnostic editor brief template I’m sending my editor on day one, a thumbnail A/B testing protocol, and a feedback loop where I paste my video performance back into the skill every week so it gets sharper over time. It also generated a polished webpage version of the entire system that I can bookmark and re-read.

The whole thing … audit, leverage map, skill, editor brief, bonus pieces, the wrap-up webpag, landed in about 90 minutes of back-and-forth.

Subscribe now

What’s Quietly Profound About This

Six months ago this kind of workflow would have meant hiring a YouTube strategist, then a content writer, then a trends analyst. Or carving out a week to figure it all out yourself. Or following yet another guru’s $999 course.

What I actually did was open a chat window, describe what I was stuck on, and let the AI do the work. I made the creative calls. The audit was real research, not generic advice. The skill is mine … it lives in my account, runs forever, and gets sharper the more I feed it performance data.

The shift isn’t “AI helps me think.” It’s “AI builds the thing.” And the thing keeps working after the chat ends.

The Honest Trade-Offs

I owe you the trade-offs because Hyperagent is not free.

The 90-minute session burned $41 in API credits. That sounds like a lot for one conversation, and it is … but Hyperagent is doing real work. It opened browsers. It scraped multiple sites, even watched videos on my behalf! It generated artifacts. It updated a working doc throughout. The cost reflects actual compute, not chat tokens.

For comparison, I’d have paid more than $41 for an hour with a YouTube strategist who’d have given me less than what I got. But still … go in with eyes open. Big builds chew through credits. If you’re just dabbling, you’ll burn a starter balance faster than you think.

The other trade-off: it’s powerful enough that you can absolutely waste credits asking it dumb questions. Use it for the heavy lifts … researching, building, generating. Don’t use it for things you’d ask Claude or ChatGPT about for free.

The Codes

Now the good news.

Hyperagent is awarding $1000 in free credits to early users to actually try building stuff with it and they’ve given me 20 codes to share with subscribers of this newsletter. Each one is enough to comfortably build something real.

https://hyperagent.com/refer/KZGYSWFT

First-come-first-served until they’re gone.

I’m genuinely curious what you’ll build. The point of these tools isn’t replacing the work … it’s expanding what one person can ship in an afternoon.

Cheers, Jagger

If this made you think of someone who’s been “going to fix” their workflow for the past year, send it to them.

It’s on my Mac. At home.

I didn't have it on me - which meant paying $3 to print a page at the counter anyway. But first, I had to actually get the document.

So I pulled out my phone, opened Claude’s Dispatch tab, and typed: “Find my visa docs as I need a few of them.”

Thirty seconds later, Claude had searched my Documents, Desktop, and Downloads, found a Visa folder sitting right on my Desktop, and listed everything in it. Passport scan. VEVO check. IMMI bridging visa grant notification. Invitation Letter. The full set.

I typed: “Email me my invitation letter.”

Claude used Mac Mail, attached the PDF from my Desktop, and sent it to my Gmail. While I was standing there. While the officer was still processing my other documents.

Here’s what Dispatch actually is, because I think a lot of people are sleeping on it.

It’s a feature in Claude that lets you control your Mac from your phone - like computer use, but for your actual desktop. You leave a session running at home, and from anywhere in the world, you can ask Claude to navigate your machine, find files, open apps, click things, send emails. Real actions on your real computer.

It’s not screen sharing where you’re squinting at a tiny mirror of your desktop. You just... describe what you need, and Claude does it.

There’s something quietly profound about this that I keep thinking about.

AI has been good at generating things - text, images, ideas. That’s impressive but still kind of abstract. What Dispatch represents is AI that retrieves things. Things that already exist in your actual life. Your files. Your emails. Your calendar. Your applications.

The gap between “I need X” and “I have X” collapsing in real time … that’s where AI gets genuinely useful for normal people doing normal things.

Not writing blog posts. Not summarising PDFs. But finding the document you need while you’re sweating in a government office.

If you want to try this yourself, here’s the setup:

1. Download Claude for desktop (Mac for now)

2. Enable Dispatch in settings - it keeps a persistent session running in the background

3. Download Claude on your phone and navigate to the Dispatch tab

4. Leave your Mac on when you leave the house

That’s it. You now have an AI that can reach into your home computer from wherever you are.

Start with something low-stakes - ask it to find a file, check something in a browser, pull up an email. Get a feel for it. Then you’ll naturally start remembering it exists in the moments that actually matter.

Like a visa office on a Tuesday.

If this made you think of someone who’s always scrambling to find documents at the worst possible time, send it to them.

Cheers, Jagger

Not gently. Not diplomatically. Bluntly.

Jensen Huang - the man whose chips are inside basically every AI system on earth - was asked a simple question: if AI is supposed to make everyone more productive, why are companies laying people off?

His answer: “Because you’re out of imagination.”

Let’s call it what it really is - the CEO of Nvidia just called out his own biggest customers on national television. Meta. Amazon. Microsoft. Companies spending tens of billions on his hardware, while simultaneously cutting tens of thousands of jobs and citing AI as the reason.

Meta is reportedly cutting 15,000 people — 20% of its workforce — while doubling its AI budget to $135 billion this year. Amazon eliminated 16,000 corporate roles in January and explicitly named AI as the efficiency engine behind the decision. Microsoft cut more than 15,000 positions across 2025 while committing $80 billion to AI infrastructure.

These are the companies Jensen is talking about. His biggest customers. And he went on TV and said they lack imagination.

That’s a gutsy thing to say. It’s also not the whole story.

Here’s the thing nobody’s talking about: Jensen’s framing is inspiring and probably true at the company level. It’s a much harder sell at the human level.

His argument goes like this - when a company gains new capability, the right move is to ask what you can now build that was previously impossible. What new market can you enter? What product can now exist? The ambitious answer to “we can do more with less” is do more - not employ less.

And he’s right that some companies will do exactly that. The ones that use AI to expand instead of contract will probably eat the ones that don’t.

But Jensen’s argument assumes that the workers being cut will land somewhere in this expanding vision. That the 16,000 people who left Amazon in January are going to become AI-era architects of something new. That the answer to “I just lost my job to automation” is “don’t worry, a more imaginative company will find you.”

That’s not imagination. That’s optimism wearing imagination’s clothes.

Subscribe now

Dario Amodei - the CEO of Anthropic, the company that built Claude - has been more direct about what’s actually coming. He’s warned that AI could eliminate 50% of entry-level white-collar jobs within five years. Not because companies lack imagination. Because the economics are straightforward.

The Dallas Federal Reserve already has data showing workers aged 20–30 in AI-exposed roles have seen unemployment rise nearly three percentage points since early 2025. That’s a real shift, in a short timeframe, concentrated in exactly the people entering careers right now.

So where does that leave you?

If you’re a company leader, Jensen’s framing is the one worth holding onto. The organisations that use AI to shrink will hit an efficiency ceiling fast. The ones that use it to expand - new products, new markets, shorter cycles - will compound in ways that are genuinely hard to replicate. The technology is identical. The difference is what you’re trying to build.

If you’re an individual, the honest answer is messier. Waiting to be redeployed into an imaginative company’s expanded vision is not a strategy. The uncomfortable truth is that the burden of navigating this transition is falling on people, not organisations.

That’s not a reason to panic. But it is a reason to move.

If you’re not already building fluency with the tools that are reshaping your field, start this week. Not a course. Not a YouTube rabbit hole. Pick one task you do regularly at work and spend an hour trying to do it with AI. See what happens. That friction - the moments where it works and the moments where it doesn’t - is more valuable than any explainer article.

The people who come out of this period well won’t be the ones who had the most optimistic boss. They’ll be the ones who stopped waiting to be told what to do with the new tools and just started doing things.

Jensen’s right that imagination is the variable. He just forgot to mention whose imagination matters most right now.

It’s yours.

If you found this useful, share it with someone who’s been asking whether their job is safe.

Cheers, Jagger

Subscribe now

The moment I saw it, I thought: that should be Precious.

Precious is my 6-year-old Samoyed. She’s an enormous white cloud of fur who’s cute as anything, full of sass, quite clumsy, and deeply unimpressed by most things... I wanted her on my dock. I wanted to click her and chat with her. I had no idea how to build that. I did it anyway.

Here’s exactly how, so you can build yours too.

---

Step 1: Get Inspired (and Steal Shamelessly)

Go look at Lil Agents on GitHub: https://github.com/ryanstephen/lil-agents

It’s the reference for this whole project. Ryan built it with animated characters that walk along your dock and open a streaming chat window when clicked. The code handles all the hard macOS stuff - window positioning, transparent layers, dock integration. Rather than starting from scratch, I used it as the blueprint and built something inspired by the approach. Thank you Ryan!

You don’t need to fork it or understand Swift deeply. You just need to understand what it does so you can describe what you want to an AI and have it build your version.

---

Step 2: Create Your Pixel Art with Nano Banana Pro

This is where most people would get stuck. Character art is hard. Consistency across multiple poses is even harder.

I tried a few AI image tools and Nano Banana Pro was the clear winner - specifically because of character consistency. Here’s the problem with most AI image generators: generate your dog in a walking pose, then generate her sitting, and you’ll get two different dogs. Different shading, different proportions, different face. It looks terrible in a sprite animation.

Nano Banana Pro locks onto the character. Once it has Precious - her fluffy ears, her curl of a tail, her general vibe - it keeps her consistent across every frame you generate. That’s the thing that makes this actually work.

Here’s what I generated and roughly what to prompt:

Idle / Standing:

“Pixel art Samoyed dog, fluffy white fur, black outline, side view facing right, standing idle pose, transparent background, 64x64 sprite, cute chibi style”

Walk Cycle (4 frames):

“Pixel art Samoyed dog sprite sheet, 4-frame walk cycle, 2x2 grid, side view facing right, fluffy white, transparent background, 64x64 per frame, same character as reference image”

Sitting:

“Pixel art Samoyed dog sitting, same character as reference, fluffy white, side view, transparent background, 64x64”

Sleeping:

“Pixel art Samoyed dog lying down, eyes closed, curled up, fluffy white, same character, transparent background”

Happy (tail wagging):

“Pixel art Samoyed dog, excited expression, tail raised, same character, fluffy white, transparent background”

Generate your idle pose first. Then use it as a reference image in every subsequent generation - that’s how you get consistency. Save each one as a PNG with transparency. Name them: idle.png, walk_sheet.png, sit.png, sleep.png, happy.png.

---

Subscribe now

Step 3: Build the App with Claude Code

Open Claude Code and paste this prompt exactly:

---

“I want to build a macOS desktop companion app in Swift using Swift Package Manager (no Xcode needed). The app should:

- Display a transparent floating window just above the dock

- Show an animated pixel art sprite of my dog that walks back and forth

- Have 5 sprite states: walking, idle, sitting, sleeping, and happy

- The walk_sheet.png is a 2x2 grid of 4 frames - split it into individual frames in code

- Behaviour based on activity: walk/idle when recently active, sit after 20 seconds idle, sleep after 60 seconds idle

- When I click the dog, open a dark-themed chat window

- While waiting for an API response, show the sitting sprite

- When a response arrives, show the happy sprite for 2 seconds

- Connect to the Anthropic API using streaming (claude-sonnet-4-6 model)

- Ask for my API key on first launch and store it in UserDefaults

- Show a small menu bar icon with options to update the API key and quit

- Images will be in the app bundle’s Resources folder

The system prompt for the AI should be: ‘You are [your dog’s name], a friendly [breed] who lives on the user’s Mac. Be warm, helpful, and playful.’

Start with Package.swift and work through each Swift file. Build a build.sh script that compiles with swift build -c release and packages the result into a .app bundle.”

---

It will ask clarifying questions. Answer them. Run the code it gives you. When something breaks, paste the error back in and say “fix this.” Iterate. This is the loop.

It took me an afternoon. Genuinely.

---

Step 4: Drop In Your Sprites and Launch

Once your app compiles and build.sh runs successfully:

1. Right-click your .app → Show Package Contents → Contents → Resources

2. Drop in your 5 PNG files with the exact names above

3. Launch the app

4. Enter your Anthropic API key when prompted (get one at console.anthropic.com)

5. Your dog appears above the dock (mine is hidden)

That’s it.

---

What This Actually Shows

There’s a bigger point here that goes beyond building a desktop pet.

Six months ago this project required knowing Swift, AppKit, sprite animation, API integration, and macOS window management. It would’ve taken a developer a few days. The pixel art would’ve taken an artist.

Now it took one afternoon and two tools: Nano Banana Pro for the art, Claude Code for the build. I described what I wanted. I iterated when things broke. I made design decisions - Precious’s animations, her personality in the system prompt, what triggers each state. The creative and design judgment was mine. The execution was AI.

Precious is on my dock right now. She just sat down, which means I’ve been ignoring her for twenty seconds. She’ll go to sleep soon if I don’t click her.

I find this unreasonably delightful.

If you found this useful, share it with someone who’s been waiting until they “know enough to build something.”

Cheers, Jagger

Subscribe now

The trick? A tool called Ollama. It lets you run open-source language models on your own computer, and Claude Code can plug straight into it. No API key. No monthly bill. No cloud.

Here’s how to set it up in about 10 minutes.

What’s Actually Happening Here

Claude Code is the tool - it’s the interface you type into, the thing that writes and edits code for you. Normally it connects to Anthropic’s servers in the cloud, which costs money.

Ollama is like a mini AI server that runs on your laptop. It downloads an open-source model and serves it up locally. So instead of Claude Code talking to the cloud, it talks to Ollama sitting right there on your machine.

Same experience. Zero cost. The trade-off? Your computer does the heavy lifting, so you’ll want decent RAM. And the local models aren’t quite as sharp as the cloud versions. But for learning, experimenting, and building stuff? More than good enough.

Step 1: Install Ollama

Head to the Ollama website and download it. Works on both Windows and Mac.

Step 2: Download a Model

Open up your terminal (PowerShell on Windows, Terminal on Mac) and run:

ollama run gpt-oss:(size)

Replace (size) with whichever variant suits your machine. The general rule: bigger model = smarter responses = needs more RAM. If you’ve got 16GB of RAM, you’ll be fine with the mid-sized options. If you’ve got 32GB+, go large.

You can browse all available models on the Ollama website. Pick one with a large context length - that’s what lets it handle bigger codebases without losing the plot halfway through.

Step 3: Install Claude Code

On Mac:

curl -fsSL https://claude.ai/install.sh | sh

On Windows (PowerShell):

https://claude.ai/install.ps1 | iex

Windows users - if the claude command doesn’t work after install, you might need to add it to your system path. Go to System Properties → Environment Variables → System Variables → Path → New → then paste your install directory (usually C:\Users\<your-username>\.local\bin).

After that, typing claude in your terminal should fire it up.

Step 4: Connect Claude Code to Your Local Model

This is where the magic happens. You’re telling Claude Code to stop looking at the cloud and start talking to your local Ollama instead.

The easiest way:

ollama launch claude

Then select the model you downloaded earlier. Done.

If that doesn’t work for whatever reason, you can do it manually with environment variables:

On Mac:

export ANTHROPIC_AUTH_TOKEN=ollama
export ANTHROPIC_BASE_URL=http://localhost:11434
claude --model gpt-oss:(size)

On Windows (PowerShell):

$env:ANTHROPIC_AUTH_TOKEN="ollama"
$env:ANTHROPIC_BASE_URL="http://localhost:11434"
claude --model gpt-oss:(size)

First time you run it, Claude Code will ask you to confirm that your project files are trustworthy. Say yes and you’re in.

That’s It. You’re Running Claude Code for Free.

No subscription. No API credits ticking down. No anxiety about how many prompts you’ve got left this month. Just you, your machine, and a local AI ready to build stuff.

Want to Test It? Build a Pomodoro Timer.

Seriously. Once you’re set up, paste this into Claude Code and watch what happens:

“Build a classic Snake game in Python with Pygame. Grid-based movement, arrow key controls, food spawning, score counter, increasing speed as the snake grows, wall and self-collision detection, game over screen with final score. Single file, clean code.”

If it spits out a working Pomodoro Timer, your setup is solid.

The Honest Trade-Offs

I’m not going to pretend this is identical to the paid cloud experience. It’s not. Local models are generally less capable than the latest cloud models. They can be slower depending on your hardware. And if you’re working on a massive codebase, you might hit context limits faster.

But for learning how AI-assisted coding works, for building side projects, for experimenting without worrying about cost - this is genuinely brilliant. It’s also a great way to figure out if the paid version is worth it for you before committing.

And there’s something satisfying about the whole thing running on your machine. No internet required. No data leaving your laptop. Just pure local AI vibes.

If you found this useful, send it to someone who keeps saying AI tools are too expensive to try.

That’s not my prediction. That’s the logical conclusion of what’s already happening with personal AI agents and it’s being driven by the same kind of shift that killed Blockbuster, turned newspapers into websites, and made your local travel agent obsolete.

Let me explain.

The MyFitnessPal Problem

Here’s a question: why do you need a calorie tracking app when your AI agent already knows where you are?

Think about it. Your personal agent knows your schedule. It knows you just walked into a restaurant. It knows you’ve been trying to eat cleaner this month. It knows you had a heavy meal last night. It knows your weight goals.

So when you sit down at that restaurant, your agent can already suggest what to order - without you ever opening an app, scanning a barcode, or manually logging anything.

Peter Steinberger, the creator of OpenClaw (the open-source AI agent that just hit 180,000+ GitHub stars), put it bluntly in his recent conversation with Lex Fridman: “Why do you need MyFitnessPal when the agent already knows where I am?”

He’s right. And MyFitnessPal is just the tip of the iceberg.

The Pattern Is Everywhere

Once you see it, you can’t unsee it.

Your sleep tracker app? Your agent already knows how you slept — it can pull data from your wearable, adjust your gym workout based on your recovery, and text your training partner that you’re doing a lighter session today. No app required.

Your smart home app? Steinberger described how his agent controls his cameras through their API directly. No need to open the Sonos app when your agent can talk to the speakers. No need for the Ring app when your agent already has access.

Your calendar app? “I don’t wanna open a calendar app,” Steinberger said. “I just wanna tell my agent, ‘Hey, remind me about this dinner tomorrow night,’ and maybe invite two of my friends and send a WhatsApp message to them.” One sentence replaces three apps.

Your Uber Eats app? Your agent can figure out how to use your phone, open the browser, navigate the site, and place the order. Or call an API directly. Or find a better deal elsewhere. The app becomes invisible — just a slow API your agent calls on your behalf.

Every App Is Now Just a Slow API

This is the key insight that most people are missing.

It doesn’t matter whether a company wants to be agent-friendly or not. If your AI agent can access something through a browser, it’s already an API. A slow one, sure. But a working one.

Steinberger watched his agent happily click the “I’m not a robot” button on websites. Cloudflare and others are trying to block bots, but the reality is straightforward: if a human can access it through a browser, an agent can too. It’s just a matter of speed.

Companies that get ahead of this — that build proper APIs and make themselves easy for agents to work with — will thrive. They become the preferred, fast path.

Companies that fight it? They just become the slow path. And eventually, users stop caring which path their agent takes.

Who Dies, Who Survives

Here’s the split as I see it:

Dead apps walking: Single-purpose utility apps that exist to display or capture simple data. Calorie trackers, habit trackers, basic to-do lists, weather apps, simple finance dashboards, most smart home controllers. Anything where the core value is “show me this data” or “let me input this data.” An agent does both better because it has context these apps never will.

Apps that transform: Companies that rapidly pivot to being agent-facing services. Uber Eats doesn’t die — it becomes the delivery API your agent calls. Spotify doesn’t die — it becomes the music service your agent manages. The app layer disappears, but the underlying service survives. The companies that build clean, agent-friendly interfaces fastest will win this transition.

Apps that survive: Anything where the core value is the experience itself, not the utility. Games. Creative tools where you want to be hands-on. Social platforms (though even these will change dramatically). Anything where removing the human interaction removes the point.

What This Means for You

If you’re a builder or developer: start thinking about how your product works when no human ever opens your UI. What’s your API story? What’s your agent story? Because the companies that answer those questions first will eat the lunch of those that don’t.

If you’re a user: this is genuinely exciting. The subscription fatigue, the app clutter, the constant context-switching between 15 different tools to get one thing done — that era is ending. Your agent handles the mess. You just say what you need.

If you’re an investor: pay very close attention to which companies are building for the agent-first world versus which are still building for the app-store world. That distinction is about to matter a lot.

The age of the app isn’t ending overnight. But the age of the app as the primary interface between you and digital services is absolutely ending. Your agent is becoming your operating system.

And that operating system doesn’t need 80 apps. It needs APIs, data, and permission to act on your behalf.

That’s it for this week. If you found this useful, share it with someone who’s still manually logging their calories.

Formula 1 coming back to Melbourne always hits different.

It’s the first race of the season.
Fresh cars.
New regulations.
Everyone saying they’ve “nailed it” until Friday practice proves otherwise.

This year feels bigger than most.

Regulation changes don’t just reshuffle the grid - they expose which teams actually understand the systems underneath. And that’s why I’m more excited than usual.

Because Formula 1 isn’t just resetting aerodynamics or car concepts.

It’s doubling down on AI.

Formula 1 Is No Longer Won on Track Alone

We still like to believe races are decided by bravery under braking or raw driver talent.

That’s part of it. But it’s no longer the whole story.

Modern Formula 1 is about decision systems.

• Which setup to bring

• How aggressively to run the tyres

• When to pit … and when not to

• How to react when the race doesn’t follow the plan

Those decisions are now shaped - heavily - by AI-driven simulation and modelling.

Teams run thousands of simulations before a race weekend even starts. Not just for strategy, but for car setup, performance trade-offs, and failure scenarios. AI helps narrow an almost infinite set of possibilities down to a handful worth trying.

This isn’t “AI replaces engineers”.

It’s AI deciding where engineers should focus.

AI Simulation in Formula 1: What Teams Actually Use It For

When people hear “AI in F1,” they often think it’s abstract.

It’s not.

AI is already embedded in:

• Simulation workflows

• Strategy modelling

• Engineering decision loops

• Reliability and performance trade-offs

The goal isn’t perfection. It’s speed.

Speed of learning.
Speed of narrowing options.
Speed of making the least-worst decision under pressure.

What I Saw First-Hand at the Melbourne Grand Prix

Last year around the Australian Grand Prix, I attended a Dell x McLaren session where Dan Keyworth, Head of Business Technology at McLaren Racing, walked through how McLaren uses AI and data across the team.

What stood out wasn’t the tech buzzwords.

It was how operational everything was.

AI wasn’t a future roadmap item.
It was already part of how the team functioned day to day.

McLaren talked about AI the same way teams talk about hydraulics or telemetry.

Not optional.
Just part of the machine.

And you could see it working.

Results Don’t Lie

Fast forward to the season results:

• McLaren won the Constructors’ Championship

• Lando Norris won the Drivers’ Championship
  (though I’d be lying if I said I wasn’t rooting for hometown boy Oscar Piastri)

That doesn’t happen by accident.

AI Is No Longer “Quietly” Reshaping Formula 1

People like to say AI is quietly reshaping Formula 1.

I don’t think that’s true anymore.

It’s not quiet.
It’s central.

The difference is most fans only see the output … not the system.

They see:

• A perfect undercut

• A bold tyre call

• A strategy that “just worked”

What they don’t see:

• The models that ran overnight

• The simulations that killed bad options before the race began

• The decision systems ranking scenarios in real time while cars fly past at 300km/h

AI has become the backbone of how F1 teams think.

To prove this point, just a few days ago, Mercedes announced a partnership with Microsoft, where they will aim to turn data into insights, smart decisions and ultimately performance on the track. AI plays a huge role in this.

Why Regulation Changes Make AI Even More Important

Every time Formula 1 introduces regulation changes, uncertainty spikes.

That’s when AI matters most.

When nobody knows what “optimal” looks like yet, the teams with the best simulation, data infrastructure, and decision-making loops adapt faster.

They:

• Test more ideas digitally

• Fail cheaper

• Converge on better solutions sooner

With Melbourne as the first race, we’re about to see who actually understands their systems - and who’s still guessing.

Why Most Fans Still Underestimate AI in Formula 1

Ask the average fan about AI in F1 and you’ll hear:

• TV graphics

• Strategy predictions

• Cool dashboards

That’s the shallow layer.

The real impact is upstream:

• Design decisions made months earlier

• Strategy frameworks built long before lights out

• Operational resilience when things go wrong

Formula 1 teams now operate more like high-performance software organisations than traditional racing teams.

AI is the connective tissue.

Why This Story Matters (and Why I’m Telling It)

I’m a massive Formula 1 fan.
I’m also deep in the AI space.

What excites me about Formula 1 right now isn’t just the racing - it’s how clearly it shows where AI actually adds value.

As the cars line up in Melbourne on 8 March 2026, with new regulations and fresh uncertainty, we’re not just watching a race.

We’re watching the output of thousands of AI-driven decisions made long before the engines fire.

Most people have no idea how much of the sport now runs on that invisible layer.

That’s what makes this season fascinating.

I can’t wait.

—
Jagger

If this was useful, subscribe to AI the boring.

Subscribe now

I can’t use Sora yet. It’s only live in the US and Canada.
But over the past few days, I’ve watched enough clips to see what’s coming and it’s wild.

Not because the tech is magic (it’s not).
But because it might be the first AI product that feels social instead of smart.

The first video that stopped me scrolling

I saw a short clip of someone walking through a rainy street, talking casually to the camera. The reflections on the pavement looked real. The person’s face, the way the jacket moved … all believable.

It wasn’t filmed. It was generated from text.

Then I noticed something else: every comment was someone tagging a friend saying, “Make one of you doing this.”

That’s when it clicked … Sora isn’t just an AI video tool. It’s an interaction machine.

From “tools” to “toys”

Most AI products are built for utility … they help you write, code, summarise, automate.
Useful, sure, but soulless.

Sora is different. It’s designed for play.
It’s closer to TikTok than ChatGPT. You don’t go there to get something done. You go to see what happens.

That’s a shift that matters.

AI so far has been single-player: you prompt, it outputs.
Sora looks like the first multi-player experience.

You can include people you know, remix their clips, or make versions of shared jokes. It’s not about creating, it’s about participating.

That’s the ingredient that made TikTok take off … not the algorithm, but the feeling that everyone’s in on the same bit.

Watching from the sidelines

Being in Australia, I can’t try it yet.
But it’s interesting to see what people are doing when the tech is still limited.

They’re not pushing for realism. They’re testing boundaries.
“How far can I push this prompt?”
“Can I make a scene that feels slightly wrong?”
“Can I make something that looks like my memory of an event, not the event itself?”

It’s playful in a way AI hasn’t been in a long time.

That’s what scares me (and excites me). Once it lands here, everyone’s going to flood their feeds with clips. Not because they want views, but because it’s frictionless fun.

Subscribe now

The hidden shift

If you zoom out, Sora isn’t about deepfakes or content creation. It’s about attention.

TikTok owns short-form human attention.
Sora could own synthetic attention … infinite, personalised, self-referential media.

That’s not a claim that it’ll kill TikTok overnight. But the pattern is familiar:

• Lower barrier to creation.

• Built-in remix culture.

• Algorithmic distribution that rewards experimentation.

The more people use it, the better it gets.
And the better it gets, the more we’ll start trusting synthetic media as normal.

My take

This isn’t “the next big thing.” It’s the next inevitable thing.

AI tools started as productivity software.
Now they’re becoming culture engines.

And when the culture moves, the money follows.

For now, I’m happy to sit back and watch the chaos unfold from here in Melbourne.
But I don’t think it’s hyperbole to say: once this rolls out globally, every creator, brand, and even meme account will have to rethink what “video” means.

Because when everyone can generate reality on demand, the only thing left that’s real … is taste.

Subscribe if you haven’t.

Subscribe now

- Jagger

This is the go-to newsletter and community for no-code AI tools, news and productivity insights.

Did you know that the most in-demand tech skill of 2025 isn’t Python or JavaScript-it’s Vibe Coding.

Let that sink in.

The biggest opportunities in tech aren’t going to programmers anymore-they’re going to Vibe Coders.

The best thing... this course works for Absolute Beginners!

You don't need to have any prior experience...

• Zero coding-just drag, prompt, and deploy

• Step-by-step videos (no assumed knowledge)

• Lifetime access + updates (AI moves fast-we keep you current)

ENROLL NOW

🚨 Vibe Coding = Your Career Insurance

1️⃣ Get Higher Salary 2️⃣ Future-Proof Your Career 3️⃣ Beginner-friendly 4️⃣ Upgrade Your CV

The Problem? Almost no one teaches it… until today.

👉 JOIN THE COURSE NOW (95% OFF!)

(Price increases in 4 days)

This course will be your step-by-step guide. Fully hands-on.

⏰ Deadline: This discount expires in 4 days-and likely never repeats.

ENROLL NOW

If we have not met, I’m Jagger-an AI Product Manager, AI agency founder, and published in AI space-and I’ve worked hard to create the easiest path to building tools using AI without code (=Vibe Coding).

Why This Course is Different

Most AI courses teach theory. We give you the exact tools & skills to:

• Launch AI products (like his agency’s client apps)

• Automate workflows (tested in real businesses)

• Solve problems fast (no coding, no guesswork)

P.S. In 2025, there will be two types of people: Those who use AI apps, and those who get paid to build them. Which side will you be on?

👉 SEE COURSE & ENROLL NOW

Check out the curriculum & free preview videos!

See you inside,

Jagger

If you found this useful, subscribe for practical AI insights without the hype. You can also connect with me on YouTube, X, Instagram & LinkedIn.

This is the go-to newsletter and community for no-code AI tools, news and productivity insights.

TL;DR: Traditional SEO is losing ground to AI-powered search. If ChatGPT, Claude, Gemini, or Perplexity aren’t referencing your brand when people ask questions… you don’t exist. This is your crash course in LLM Engine Optimization (LEO).

🚨 Want Help Rewriting Your Content for LEO?

If you want to turn your blog post, landing page, or newsletter into something ChatGPT might actually cite, I built a free GPT that does just that.

ChatGPT - LLM Optimization Rewriter

Rewrite your content so it gets cited by AI tools like ChatGPT and Perplexity.

chatgpt.com/g/g-685215ba6304819187eeac4a6e0af593-llm-optimization-rewriter

What Is LEO and Why It’s a Big Deal

LLM Engine Optimization is the process of shaping your content so it gets used by language models - not just indexed by Google. The game has changed:

• SEO = ranked links

• LEO = cited answers

AI models don’t send people to your site. They quote you (or they don’t). That means visibility = citation.

And citation means influence, trust, and traffic from AI-native users.

How AI Models Choose What to Cite

When someone asks ChatGPT or Perplexity a question, here’s what happens behind the scenes:

Signal

What the model looks for

Authority

Are you mentioned across credible sources?

Comprehensiveness

Can one of your pages answer the whole question?

Freshness

Is your info current and regularly updated?

Context Match

Does your paragraph directly address the query?

Sentiment

Do other sources reference you positively?

It’s not just about what’s on your site - it’s about your reputation across the internet.

And here’s the kicker: AI tools usually only reference 1–3 sources. So you either make the cut or you don’t show up at all.

How To Structure Your Newsletter for LLM Discovery

Here’s how to future-proof every issue:

1. Start Strong

Put the most important insight in the first paragraph. Think: clear answer first, context second.

✅ “LLM Optimization helps you show up in AI answers. Here’s how to actually do it, step-by-step.”
❌ “AI is changing how we search… and that has some implications…”

2. Use Question-Based Subheadings

Match how people ask questions. For example:

• ❌ “SEO vs LEO”

• ✅ “What’s the difference between SEO and LLM Engine Optimization?”

3. Make Each Section Stand Alone

Every paragraph should work like a mini answer box. No rambling intros - just answer the question and move on.

Good structure:
H2: What Is LLM Optimization?
One-paragraph definition
Bullet points or examples
Optional: link to related post

4. Add a Table of Contents

If your post is over 1,000 words, use a clickable table of contents. This helps LLMs parse your structure and improves user navigation.

Example: Before vs After

❌ Original Newsletter Snippet:

“Let’s talk about AI search. Google’s new SGE feature is rolling out, and it’s changing how we think about content visibility…”

✅ Optimized for LEO:

What Is Google SGE and Why Does It Matter for Content Creators?
Google’s Search Generative Experience (SGE) uses AI to generate answers directly in search results. Instead of showing 10 blue links, it summarizes from trusted sources. If your brand isn’t in that answer box, you’re invisible.

Example:

Bonus Tips for LEO-Optimized Content

• Answer real user questions - use tools like Perplexity, AnswerThePublic, and “People Also Ask” to find phrasing

• Structure your insights like Wikipedia entries: answer > explain > elaborate

• Sprinkle first-party data or frameworks throughout (LLMs love citing original thinking)

• Add “Last updated” dates and keep evergreen content fresh

• Link to trusted sources and get linked from high-signal blogs, roundups, and communities

• Review your server logs to monitor the access patterns of search engines and AI crawlers, such as GPTBot or BingBot, on your website.

What to Do Next (Your LEO Sprint)

Week

Action

1

Run your newsletter through ChatGPT or Perplexity - are you cited?

2

Rewrite your most-read issue with question-style headers

3

Add a definitive “pillar” post to your site (3,000+ words, cited sources)

4

Seed mentions: get featured in a roundup, podcast, or expert quote list

Final Thought

If your name isn’t coming up in AI answers, it’s not because your content sucks. It’s because it isn’t optimized for how AI thinks.

Start with structure. Focus on questions. Create source-ready insights.

Your future traffic will thank you.

If this helped, subscribe to AI the Boring - your no-hype guide to working smarter with AI. Or forward this to someone still obsessing over Google rankings. They’ll thank you in six months. You can also connect with me on YouTube, X, Instagram & LinkedIn.

🔥 Just Dropped: New YouTube Video

Quick heads-up - my latest YouTube video is live now! Check it out and let me know what you think in the comments!

The SMARTEST Way to Build AI Apps in 2025 (No Code + MCP)

Subscribe

This is the go-to newsletter and community for no-code AI tools, news and productivity insights.

Let’s cut through the noise.

If you still don’t have ChatGPT on your phone, what are you doing?

No seriously - what are you actually doing? I’ve heard every excuse by now: “I don’t trust AI,” “I’m too busy,” “I’m waiting to see where it goes.” But let’s call it what it really is: you’re either lazy, or you’re hoping this whole AI thing just blows over.

It won’t.

AI isn’t the future. It’s the right now. Just ask the company in Portugal that launched an AI chatbot and raked in $100 million in sales. Not projections. Not potential. Real results.

So… what’s the difference between them and everyone else?

They actually used the tools.

AI Doesn’t Care About Your Opinion

Here’s the truth: AI doesn’t need your approval. It’s not waiting for you to catch up. It’s already reshaping industries - marketing, real estate, logistics, law, finance. And if you’re sitting on the sidelines because it feels uncomfortable or overwhelming, you're not being cautious. You’re being left behind.

That’s harsh, but it's real. Most people want change without the discomfort of change. They want the benefits of AI without the icky learning curve. But the gap between early adopters and everyone else is widening fast.

You either build with it or get replaced by someone who does.

Your 50-100 Hour Reality Check

Want to know how to stop falling behind? It’s not about buying the right software or finding a unicorn AI employee.

It's about time.

The only real requirement is this: invest 50-100 hours into understanding what AI can do for you, your career, and your business.

Here’s a dead-simple plan to get started:

1. Download ChatGPT (or Claude or Perplexity…just pick one)
   Put it on your phone. Open it daily. Use it for real tasks, not just “fun” prompts.

2. Ask better questions
   Try things like:

   • “How will AI affect real estate?”

   • “What are 5 ways AI can automate customer support for a small business?”

   • “Write me an email pitch for [insert your product].”

3. Build a basic AI workflow
   Pick one repeatable task in your week - writing content, analyzing reports, summarizing meetings - and replace it with AI. Not tomorrow. Today.

You don’t need to understand the backend or write a single line of code. You just need to get in the arena. Use the tools. See what’s possible.

Still Think It’s Optional?

Let’s zoom out.

AI isn’t just doing “tasks.” It’s shaping preferences, decisions, and distribution. Here’s what’s coming:

• You say, “Find me a good local café with oat milk lattes and a quiet vibe to work for a few hours.”

• Your AI checks real-time noise levels, Wi-Fi quality, and menu options - then books you a seat and sends directions.

• You didn’t scroll Google Maps. You didn’t read reviews.

• If your café isn’t known, tagged, or optimized for discovery - you don’t even make the list.

That’s the world we’re heading into. AI will decide, and we’ll thank it for the convenience.

Unless you build brand, build presence, and build systems that play with AI - not against it- you won’t just miss out. You’ll disappear.

Last Excuse: “I’m Too Old”

I hear this one a lot:
“I’m 60, I’ll just ride it out.”
Respectfully…no, you won’t.

Even if you’re not working, AI will impact your life. Your health. Your bills. Your shopping. The sooner you understand how it works, the easier your life will be. Tech doesn’t care about your age. But you should care about your leverage.

Do the Work - Or Get Worked

Look, none of this is about fear-mongering. It’s about urgency. Not panic. Just momentum.

Nobody’s asking you to become an AI engineer. But if you keep avoiding the tools that are rewriting the rules - you’re not protecting yourself. You’re making yourself replaceable.

Don’t be the person saying “I should’ve learned this five years ago.”

Be the one who says “I saw it coming and I adapted.”

You’ve got two things to do this week:

1. Download ChatGPT and start playing.

2. Block out your first 5 hours of AI research.

Then keep going. Stack hours. Build skills. Build leverage.

The train is moving. Time to stop watching it go by - and start driving it.

Cheers, Jagger

If you found this useful, subscribe to AI the boring for more practical AI insights without the hype. You can also connect with me on YouTube, X, Instagram & LinkedIn.

🔥 Just Dropped: New YouTube Video

Quick heads-up - my latest YouTube video is live now! The video is a beginner-friendly tutorial on building AI-powered apps and automation using Lovable (a no-code app builder) and n8n (an open-source automation tool). It walks viewers through creating a complete no-code web app and automation workflow, specifically building an AI Personal Trainer app by integrating tools like GPT-4o or Claude 3.7 with webhooks and AI agents, all without needing coding experience.

Check it out and let me know what you think in the comments!

Subscribe

This is the go-to newsletter and community for no-code AI tools, news and productivity insights.

It's wild how quickly things change in tech.

Just months ago, businesses were panicking about AI models ingesting their precious website content. Today? The concern has completely flipped. Now the real fear is being invisible to AI systems like ChatGPT.

Let me break down why this matters more than most realize.

The Quiet Revolution in How People Find Things

Traditional search engines give you a digital scavenger hunt. You get links. You click. You scan. You go back. You click another link. Maybe you find what you need after several attempts.

But AI assistants? They skip all that. Ask ChatGPT where to find the best coffee in Melbourne, and it doesn't send you on a link-clicking expedition - it just tells you. Directly. With reasoning.

And here's the thing I've noticed recently: I'm not alone in this shift. My browser history shows a clear pattern - I'm checking Google less and asking ChatGPT more. Every week, the ratio tilts further.

The Shopping Game-Changer

OpenAI's recent shopping integration isn't just a feature update - it's a fundamental shift in how people discover and buy products.

When a user asks about running shoes for flat feet, ChatGPT doesn't just explain what features to look for - it now displays actual products with visuals, descriptions, and links to purchase.

This is massive. Think about it:

• No paid ads cluttering the results

• No SEO games determining visibility

• Just AI-determined relevance to the user's specific situation

The businesses that understand this new reality first will have an enormous advantage.

Your Website Isn't Your Front Door Anymore

Here's the uncomfortable truth: for a growing number of people, your carefully crafted website isn't their first touchpoint with your business.

Their first impression comes from how you're represented inside AI tools like ChatGPT, Claude, or Perplexity.

If you run a beach resort in Bali, your potential customers aren't just Googling "best Bali beach resorts" anymore. They're asking AI assistants about their specific travel needs:

"I want a quiet beach resort in Bali with great food, a spa, and activities for kids."

The AI gives them tailored recommendations based on their criteria. And critically - some resorts will make the list, and others won't.

The Transaction Gap is the Real Opportunity

Let's talk about what happens next.

After ChatGPT recommends that beach resort, the user has two options:

1. Click through to the resort's website, navigate booking pages, enter credit card details, etc.

2. Or simply say, "That sounds perfect. Can you book it for me?"

Right now, most businesses are only set up for option #1. But option #2 is quickly becoming possible - and for users, it's infinitely more convenient.

The businesses that bridge this transaction gap first will have a serious competitive edge. When five comparable options are presented, but only yours can be booked within the conversation flow, guess who wins the business?

Breaking Down How AI Decides Who to Recommend

Understanding how you get recommended by AI is critical. While the exact algorithms are proprietary, we know some key factors:

1. Structured data clarity - How well can AI systems understand exactly what you offer? Businesses with clear, structured data win here.

2. Social proof verification - AI systems look for consistent positive mentions across the web. Are customers talking about you favorably?

3. Specificity advantage - Generic businesses get lost. Those with clear specializations and unique value propositions stand out in AI recommendations.

4. Transactional readiness - As platforms enable direct bookings and purchases, having the technical integration ready will be critical.

Local Businesses Need This Most

This shift affects everyone, but local businesses have the most to gain (or lose).

Take our Bali resort example from earlier. When a tourist asks ChatGPT about the best resort in Bali, appearing in that recommendation list means everything. Being left off could be existential.

The good news is that local businesses can often move faster than corporate giants. While larger companies navigate bureaucracy to adapt, nimble local operations can quickly optimize for AI visibility.

Four Steps to Make Your Business AI-Visible

Here's my action plan for making your business visible in the new AI-first world:

Step 1: Audit Your Digital Footprint

Run experiments now. Ask ChatGPT and other AI tools about your business category, location, and specific offerings. See if you appear, and what information is presented. This is your baseline.

Step 2: Structure Your Online Information

AI systems love structure. Implement schema markup on your website. Ensure your Google Business Profile is complete. Make your offerings crystal clear across all platforms.

Step 3: Build a Consistent Review Strategy

Reviews matter tremendously in how AI tools assess your business. Implement a systematic approach to gathering authentic customer feedback across platforms.

Step 4: Prepare Your Transaction Layer

Even if direct AI transactions aren't fully available in your industry yet, start preparing. Streamline your booking/purchase processes and explore API integrations that will enable seamless AI-initiated transactions.

The Time Advantage Is Massive

The best part about this shift? We're still early.

While everyone else is focused on traditional marketing channels, you have a rare opportunity to establish dominance in the AI recommendation space.

Those who wait until this becomes obvious will be playing catch-up for years. The businesses that become AI-visible now will build compounding advantages as these platforms grow.

The Bottom Line

The question isn't whether AI tools will become major discovery and transaction channels - that's already happening. The question is whether your business will be visible and actionable within them.

This isn't just another marketing channel to add to your mix. It's a fundamental shift in how people find and choose businesses. Miss this, and you might find yourself invisible to an entire generation of customers who start and finish their purchase journeys within AI interfaces.

What are you doing to make your business AI-visible? Hit reply and let me know!

Cheers, Jagger

If you found this useful, subscribe to AI the boring for more practical AI insights without the hype. You can also connect with me on YouTube, X, Instagram & LinkedIn.

🔥 Just Dropped: New YouTube Video

Quick heads-up - my latest YouTube video is live now! The video is a beginner-friendly tutorial on building AI-powered apps and automation using Lovable (a no-code app builder) and n8n (an open-source automation tool). It walks viewers through creating a complete no-code web app and automation workflow, specifically building an AI Personal Trainer app by integrating tools like GPT-4o or Claude 3.7 with webhooks and AI agents, all without needing coding experience.

Check it out and let me know what you think in the comments!

This is the go-to newsletter and community for no-code AI tools, news and productivity insights.

💡 OpenAI just made image generation boringly easy

OpenAI’s new model (GPT-4o) can now generate images natively. No switching models. No extra tools.

Here’s what it can do:

• Create stunning, realistic images in seconds

• Understand and edit images using natural language

• Combine text + image workflows without skipping a beat

This isn’t just fun - it’s infrastructure. Think Canva on steroids, inside ChatGPT.

Here are some examples:

1. YouTube Thumbnails

2. Advertising

3. Infographics

Greg Isenberg is a product advisor to Reddit, the CEO of Late Checkout (a startup studio), and one of the best minds online for spotting early product trends.

This week, he posted this thread on X about monetizing AI-generated images.

Here’s what stood out 👇

1. VisionBoard AI
Describe your 1-year plan:
“Launch a product, speak on stage, lose 10 lbs”
→ It turns your goals into a clean, aesthetic visual collage.
$29 per set. $50/year if you want motivational reminders.
Feels like manifestation, but for product people.

2. Airbnb Photo Enhancer (I tested this one)
Upload your average listing photos.
Get back clean, high-conversion images.
$79/property.
Smart wedge into the 7M+ Airbnb hosts market.
Could grow into full-stack hosting tools.

As mentioned, I tried out this idea and it could well be worth building.

My dining room BEFORE 4o

My dining room AFTER 4o

💭 3 more image-AI startup ideas (inspired by Greg Isenberg’s tweet) that I thought of on a walk today

3. PitchDeck Polisher
Founders upload a rough deck.
Get back a beautifully designed version, AI-enhanced and pitch-ready.
$99/deck. Templates powered by what’s trending in YC, Sequoia, etc.

4. Before & After AI
Creators upload their “before” shot—hair, fitness, skin, room makeover.
AI generates a tastefully aspirational “after” version.
Perfect for coaching funnels, ecommerce, or engagement bait.
$15/image or $40/month for batch processing.

5. Menu Makeover
Restaurants upload a photo of their printed menu.
Get a redesigned, modern version (print and digital).
Bonus: Add AI-generated food photos for each item.
$49/menu. Great wedge into SMB design-as-a-service.

That’s it for this week.
OpenAI shipped. Greg shipped. You?

Expect a new wave of image-first startups.

Cheers,
Jagger

P.S. If you think this newsletter might help someone you know, feel free to forward it along. Sometimes, the right idea starts with the right conversation.

Oh, and do not forget to enter the draw to win a Perplexity PRO membership (closing this Friday). Enter here.

and…here is my latest video on YouTube

How To Use Manus AI To INSTANTLY Generate $1M+ Startup Ideas

Connect with me here.

• YouTube

• Linkedin

• Twitter/X

• Instagram