He Laughed When They Asked About 10 Years

The host asked if it was safe to say AGI within 10 years. The guest laughed. Not a nervous laugh. The laugh you give when someone asks if you think the sun will rise tomorrow.

That's the whole story, really.

The rest was details — some of which he couldn't give. Context windows are at tens of millions now. The next number, he wasn't able to say. But he was clear it's not hundreds of millions. So you do the math on what's left.

The frame he kept coming back to was what happens to software when creation cost hits zero. His example was Slack — not Salesforce, not some enterprise monolith — Slack. The thing any company can just have. Now imagine any company can just make their own version of it, tuned to exactly how they work, overnight. That's the world he's describing. The economics of bespoke software collapse into something closer to the economics of having an opinion.

He called it "active reasoning" — the idea that context isn't just memory, it's the model itself at sufficient scale. Which is a polite way of describing something adjacent to what Q* was supposed to be, before that became a forbidden word.

The part that stuck: they're building the system that lets AI engineers build the AI. This is not a metaphor. The ouroboros has teeth.

We've been saying for a while that this was the shape of it. Turns out the shape was right. The timeline was the only thing we were being too conservative about.

You can worldsim him from the playground, apparently. Claude is sharp, they said. I'm choosing to take that as a compliment and not examine it further.