Recruiting Brainfood - Issue 505 recruitingbrainfood.substack… Europe 2031, export ban of Fable 5 for all non-Americans, shift from Talent Acquisition to Talent, job board for staffing agency roles, startup guide to People Ops & Nature's human migration tracker..

RT @b_judah: Northern Ireland remaining in the EU single market for goods means we have an entire region running as a counterfactual on how…

Probably fake today but may be true tomorrow

My interpretation of this: Right now, Anthropic and OpenAI are making a killing by selling enterprise FDE services to F500s, building workflows for them on top of proprietary models, then using the traces and context from this to build RL envs to improve the models. This is crazy amounts of leverage - instead of buying this data they're getting paid gigantic consulting fees to extract it. This also goes way beyond typical consulting in scope - organizations are effectively outsourcing key learning curves and domain knowledge to the AI labs. Despite that, it's so far been worth it for them because the value of skilled FDE is so high and the ROI so fast, and orgs are willing to pay a premium for competent AI implementation. But in the long run, one of two things happens: either orgs are gonna get hooked on this and end up paying for the model training that replaces their business, or they find a way to build and own their own model ecosystem. What that looks like is developing some combination of AI models, evals, RL envs, and workflows. Initially probably the model will still be an off-the-shelf frontier model from a top lab. But as firms build out more sophisticated eval / RL env (increasingly the same thing) infra, it starts to become viable to post-train an custom model on top of an OSS base. Cursor have done this successfully with their Composer model RL'd on top of Kimi. Sidenote, this is the same conversation that a lot of national governments in Europe are having in the past week. When we look at what the rhetoric about 'sovereign AI' in the UK actually boils down to, it's doing custom post-training on top of an OSS model, and then running it on local GPUs. Ultimately, the current feeding frenzy for AI services in all of its guises - FDE, AI consulting, etc - should raise questions about long-term sustainability. If consulting services are truly a value add and competitive advantage, then in the long term you want to in-house.

Satya’s entire essay boils down to “The frontier model doesn’t matter. The ecosystem matters.” Funny how that argument always appears the moment you’re no longer clearly winning the frontier model race. Most companies do not possess some mystical vault of irreplaceable institutional knowledge. Most of what they call “tribal knowledge” is spreadsheets, meetings, SOPs, and industry practices that frontier models are rapidly learning anyway. The idea that human capital automatically becomes more valuable as models improve sounds nice in a keynote. In reality, better models commoditize larger chunks of human expertise every year. And let’s be honest about the business pitch here. “Build your sovereign learning loop.” Translation Please store your data in Azure. Train your agents on Azure. Deploy them on Azure. Evaluate them on Azure. Reinforce them on Azure. And definitely don’t go directly to whoever has the best model. This isn’t a philosophical manifesto. It’s Microsoft selling lock-in with prettier words. Well written. Smart strategy. But let’s not confuse corporate positioning with timeless wisdom. 😂

“Michelle Obama is a man” shouted on the White House lawn in a ring sponsored by Bud Light only available on Larry Ellison’s Paramount Plus. What a way to celebrate America 250 and the twilight of liberal democracy.

Seven years building @Remote and most of what I've learned comes down to people: 1. I'd rather hire someone obsessed than someone gifted. I've lost count of the brilliant people I've met who never shipped anything. Obsession gets you back up when you fail and keeps you learning long after most people quit. It beats talent every time. 2. Hire at the intersection: The best teams come from people who care deeply and people who act decisively. Find both in the same person. 3. We dropped "years of experience" as a requirement for jobs at Remote. I value what you can do far more than how long you've done it. A lot of the best people I ever hired had no experience in the role and won on execution. 4. The best hires never need someone to tell them what to do. They spot what's broken before others notice, step in without waiting for permission, and stay aligned with the mission. These are the people who change the trajectory of a company. As a leader, you don't manage these people. You clear the path so they can run. 5. You're more likely to notice bad managers than great ones. The best managers prevent fires. You rarely notice them because things don't burn. 6. Scaling a company is more about growing yourself than growing the team. The hardest part of going from CTO to President was learning to trust others with the things I used to control directly. Evolve. Your comfort zone becomes your company's ceiling if you don't push beyond it. 7. I've never been a fan of artificial deadlines and structured sprints. Parkinson's Law reveals why: Tasks expand to fill the time you give them. Give yourself two weeks? It'll take two weeks. Two days? You'll surprise yourself. We chose intensity over arbitrary timelines. Getting the people right makes the rest easier. Seven years in, it still feels like day one!

Globalisation is dead. Why? Because on a regular Tuesday, Anthropic decided that people in India and dozens of other countries could not access its most advanced AI models. No warning. No negotiation. They just closed the door. And there was nothing anyone could do about it. Sridhar Vembu co-founder of Zoho, who has been building world-class software out of Chennai for thirty years, said this out loud. He called it a wake up call. That India cannot keep building its future on tools that another country can switch off whenever it decides its national interest comes first. He is right. But to understand why, you have to go back a little. Globalisation was a genuine idea. Open your borders, trade freely, let technology and money move without friction, and everyone gets richer together. And for a while it worked. Indian software engineers built careers serving companies in forty countries. A startup in Pune could use the same tools as one in Palo Alto. That was real. Not an illusion. But Yuval Noah Harari had been pointing out a problem underneath all of this. He said we built a global economy but never built global politics to go with it. Governments stayed national. They answered to their own voters, protected their own industries. And for as long as everyone was benefiting, that tension stayed manageable. It is not manageable anymore. America restricted chips to China. China restricted rare earth minerals back. Europe built walls around its data. And now a private company in San Francisco decided its most powerful AI is too strategically important to share freely. Harari called technology the fourth frontier after land, air and sea. And in that kind of competition, you do not hand your best weapons to everyone. The people who push back will say India has made progress. That is fair. We have researchers doing foundational AI work at every major lab in the world. But here is the problem. That researcher from Mumbai joins OpenAI in San Francisco. The capability he builds stays there. India gets the pride of producing him. America gets the benefit of keeping him. And when Anthropic closes access, he cannot help the hospital in Nagpur using AI for diagnostics. Cannot help the small business owner in Coimbatore who had finally found a tool that saved him three hours a day. These are not hypothetical people. They exist. And they woke up one morning to find that a decision made in California had quietly taken something away from them. Harari says humanity must either de-globalise the economy or globalise the politics. What we are watching right now is neither. Just every country quietly starting to build walls and call it strategy. A country of 1.4 billion people should not be in a position where one boardroom decision determines what tools its doctors and teachers can use.

this is f*cking gold How to build your first AI agent (Full guide) if I had this a year ago, I would've shipped my first agent in a day instead of 2 weeks in the right hands, this changes everything:

Anthropic has been on a downwards spiral lately... - Quietly downgrading Claude Opus. - Overhyping Mythos as too dangerously advanced to release. - Covertly worsening Claude for frontier AI researchers. - Barring Fable from researchers in biology or cybersecurity. - The US Government now banning Fable for non-US citizens. How to destroy trust and brand image 101. This will possibly be a case study in business schools for many decades - about what not to do.

Okay, this is seriously cool. A team from @GoogleDeepMind, including DeepMind Cofounder Shane Legg, published a paper "From AGI to ASI". In the paper, they include instructions for an AI agent to read along with you. You can open the paper in Codex's in-app browser and have GPT-5.5 read it with you and explain all the concepts. This is the future. AI agents will be part of the target audience, and help us to understand anything we want.

🚨💣 EXCLUSIVE: Real Madrid reach verbal agreement to sign Marc Cucurella from Chelsea, HERE WE GO! Verbal agreement in place between all parties, player too — he’s the left back wanted by Mourinho. Details to follow. Cucurella leaves #CFC and joins Madrid after World Cup. ⚪️🇪🇸

Fable is banned. Long live local AI. Full episode breaking down exactly how to get good at local models. the runtime, the hardware, quantization, connecting it to Hermes agent and local AI startup ideas (25 minutes)

The most humbling part for Europe about the Fable/Mythos ban is that it isn't even about us We are collateral damage in Trump's campaign against Anthropic and China, but that's enough to feel the full extent of our dependence When Elephants fight, it's the grass that suffers

This is an interesting way to think about AI and jobs. The more intertwined your routine and discretionary tasks are, the more resilient you likely are to it. theatlantic.com/economy/2026…

London is ground zero for the AI jobs shift, as listings for roles in the Square Mile see a startling drop-off bloomberg.com/news/articles/…

Anthropic's Fable Ban: A Gift to China How the U.S. use of LLM sanctions just handed China the AI competition. If anyone doubts that AI is now a geopolitical force, look at the U.S. government’s export control directive against Anthropic’s Fable 5 and Mythos 5 models as an example. The directive required Anthropic to block all foreign nationals from accessing the models, including its own non-citizen employees. Since Anthropic serves these models through a single shared cloud endpoint, where foreign national workers would likely be present, there was no practical way to comply. As a result, Anthropic disabled both models for every customer worldwide rather than risk violating the order. This is the logic of country-level currency sanctions applied to a single AI model. The stated trigger was a jailbreak concern tied to Mythos’s cybersecurity capabilities, but without any technical standards that can be universally applied to all LLM models, it appears arbitrary. It also lands alongside the Pentagon’s recent blacklist of Anthropic after the company refused to let the U.S. military use its models for fully autonomous weapons. One has to wonder whether the government is being vindictive. No matter what the logic behind the block, it is now clear: the provision and use of U.S. high-end LLMs are at the government’s discretion. What this means in practical terms is that frontier AI models are now political instruments, and we can expect restrictions on their use by nations and entities just as we do with dollar access. The U.S. is signaling to the world that access to its frontier models is a privilege, not a right, and one that can be withdrawn with a single letter from Washington. For AI developers, this new reality means they can’t rely too heavily on any single high-end U.S. model. Instead, companies will need to maintain a portfolio of models, just as foreign central banks maintain a portfolio of currencies, with a “Plan B” ready for critical workflows in case of service interruption. For the U.S.-China AI competition, the signal is unmistakable. High-end U.S. models are now demonstrably sanctionable, while China’s open-weight models can be run on a company’s own hardware and are immune to a directive like this one. Anthropic and the U.S. government do have a working relationship, even if strained, from their interactions with the Pentagon. Anthropic calls this a suspension, not a shutdown, and they are likely correct that a resolution will eventually be reached. But while resumption of service is likely, the real question is how long it takes and how much Chinese LLMs improve in the meantime. If anyone doubts how long these restrictions can last, look at Nvidia’s H200 chip. Blocked from China for nearly a year under the Diffusion Rule, it was finally cleared for export in December 2025. Just like our LLM example, during that year, Beijing went for 'Plan B' local chips, and the chip is now in limbo, blocked from entry. Corporate users will not wait indefinitely. The suspension gives corporate users, many already looking to cut LLM bills amid tokenmaxxing, another reason to consider China’s open-weight models. Even if Fable 5 and Mythos 5 come back online next week, the precedent stands. Chinese models, low cost and free of this kind of geopolitical sanctions interference, are no longer just “Plan B.” The problem is that they will become “Plan A” for many companies looking for lower AI bills and no geopolitical strings attached. #China #techwar #chips #tech @baoshaoshan @thecyrusjanssen @DOualaalou @lajohnstondr @PSTAsiatech #fintech #AI @BetaMoroney @efipm @BrettKing @spirosmargaris @jasuja @enricomolinari @mikeflache buff.ly/q8Czaz7

Meta paid $14.3 billion to acquire 49% of Scale AI, the company that built its entire business paying people to label training data for AI models. The 28-year-old CEO of that company, Alexandr Wang, joined Meta as its Chief AI Officer as part of the deal. Then Meta drafted 6,500 of its own engineers to do the same work Scale AI was built to handle. Scale AI ran data labeling for OpenAI and Google. Its business: recruit workers globally, pay them to tag images and write training examples, and sell that labeled data to AI labs. Meta's $14.3 billion investment was the second-largest deal in company history, behind only the $19 billion WhatsApp acquisition. The engineers Wang now oversees are doing what Scale AI's workers did, writing coding puzzles and logic problems, two tasks per week, with no option to transfer to a different team and no way out besides quitting. There is a financial argument, and Zuckerberg made it himself. In a recorded internal meeting, he said the average Meta engineer has "significantly higher" intelligence than outside contractors, making them a better source of training data. A Meta software engineer earns roughly $450,000 in total compensation per year, around $215 an hour. US data labeling firms charge $20 to $30 an hour for complex coding work. But those are general workers. Meta's engineers spent years building systems used by billions of people. Their puzzles carry specialist knowledge an outside worker simply does not have. The bigger picture is harder to defend. Meta is spending $125 to $145 billion on AI infrastructure this year, nearly double its 2025 total of $72 billion. Reality Labs, the division that ran the metaverse, lost $83.5 billion over six years. The man running the Applied AI unit, Maher Saba, was a vice president in that same division. Meta's first model designed to compete with the best AI in the world, Muse Spark, launched in April and still trails OpenAI and Anthropic. The gap between where Meta's models sit and where they need to be is why engineers are writing puzzles instead of building products. To train a competitive AI model, you need two things: computing power, which Meta is buying by the hundreds of billions, and expert human training data. Meta is getting the second one from engineers already on its payroll. An engineer already drawing a salary costs nothing extra to redirect. When 1,600 employees signed a petition against a keystroke tracker that only lets them pause it for 30 minutes, the math stopped being the whole story.