Andrew Tulloch was offered a pay package worth as much as 1.5 billion US dollars to join Meta last year. He turned it down. A few months later he joined for a smaller, undisclosed amount. This week he left Meta to go to Anthropic.

That is the headline everyone will run, and the huge number is why. A sum that large still did not do the job, and why it failed is the story worth your time.

Tulloch spent eleven years at Meta, left in 2023 for OpenAI, then co-founded a startup called Thinking Machines Lab. Meta chased him back with an offer that could reach 1.5 billion dollars once bonuses and stock were counted. He said no, joined for less, and has now moved on again. He is the third Thinking Machines co-founder to change jobs in a matter of days. Last month, Google's chief scientist left with three colleagues to start their own company.

Three moves, one pattern.

At the same time, a researcher named Jacob Coxon quit two of the richest employers in the field, Anthropic and OpenAI. He told the public they are "gambling with our lives". His post drew more than 150 million views and pulled Congress in. Senators who had ignored the industry for years started drafting rules. One of them went on television and called it "scary stuff".

Then the head of OpenAI told his own staff the company might slow down its most advanced work, and said he hoped rivals would slow down too. Back in July, more than a thousand staff across the major labs had already signed a petition asking for a way to slow AI down.

Money stopped being the deciding vote

The obvious story is that the AI talent war has gone mad and the numbers are silly. The more useful story is simpler. At the top of this field, money is no longer the thing that decides where people go.

Reason one. Meta made one of the biggest pay offers the tech industry has ever seen, and yet he refused. Reason two. The people with the loudest warnings are leaving the money on purpose, so they can speak freely from the outside. The money did not work either way. Tulloch took a smaller offer and left. Coxon quit and went public.

So what does move them? Two things. What they believe they are building, and who they get to work with. Tulloch left for a lab whose whole pitch is the work. Coxon left because he had stopped believing in the work at all. Money was on the table for both of them, and both chose something else. That is the part a business can use, because you can offer good work and good people even when you cannot offer the most money.

The people building it are nervous in public

This matters to your business, and it matters right now. The day before Tulloch announced his exit, Meta launched a personal AI agent for ordinary users. Give it access and it can send your emails, shop online for you and finish tasks on your behalf. So look at the two things happening at once. The people who understand these systems best are getting nervous in public. At the same time, those systems are being handed the keys to your inbox and your bank card.

Think about that for a minute. The architects of this technology, the people offered mad sums to build it, are warning the public and walking away, some for far less money, some for a rival. It sounds like a film plot, and it is worth asking why it is happening. There is more to this than a pay dispute, and it changes how you should use the tools. You can still use AI. Just be clear about where a human stays in charge.

One caution before the actions. These are American labs and American numbers. The lessons below are my read across for an Australian business. Treat them as a guide rather than a proven rule.

What to do about it

Stop trying to out-pay people. Meta proved that even more than a billion dollars will not hold someone who would rather build something else. A small Australian business will never beat a big lab on salary, and it does not have to. You can offer interesting work instead. You can offer good people to work alongside. Put both in the job ad, and make it a point to mention them again at the interview. That is what moved Tulloch. The money on its own did not.

Put a human check on anything an AI agent cannot undo. Let AI draft, sort, summarise and suggest all day long. Keep a person in control of the four actions you cannot reverse: sending, buying, publishing and deleting. Once an agent emails a client, places an order, posts in public or wipes a file, the action is done and you cannot take it back. That is the point where a human signs off first.

Watch where the experts go, not where the money goes. The money went to Meta, and the experts are leaving it. When you pick tools, platforms and partners, follow the people who actually understand the technology, rather than the brand with the loudest launch. Where the experts move is a free signal, and it is public.

Use the tools this week. Keep your hand on the ones you cannot undo.