The Coffee Was Flowing

Greg Isenberg posted a thread of forty-odd predictions about AI agents, the death of SaaS, agent colocation data centres, AI-native churches, and the "most asymmetric window" he's ever seen in business. He closes with: "the coffee was flowing so had to get this off my chest."

You can tell.

There's a genre of tech prediction thread that works by taking a real observation, stripping out every complication, and projecting it forward on a ruler-straight line until it sounds like science fiction. Isenberg is good at the genre. Some of his starting points are sharp — proprietary data appreciating while generic data commoditises, the agency labour-arbitrage model collapsing, niche audiences of 5,000 becoming viable businesses. These are observations that hold up.

But then the coffee kicks in and we get "one-hour companies" (idea at 9am, first customer by 10), AI agents hiring other AI agents until the org chart "looks like a serverless function," and "agent colocation" where companies pay for their AI agents to run physically near their partners' agents. That last one is my favourite. He's reinvented "putting servers close together" and framed it as a paradigm shift.

The tell is the therapy claim. He predicts AI-native therapy and support groups will serve more people than their human-led equivalents within five years, "because they're available at 3am on a Tuesday when you actually need them." Availability is not the bottleneck in mental healthcare. The therapeutic alliance — the relationship between a person and their therapist — is the single most reliable predictor of outcomes across every modality studied. Suggesting an always-on chatbot will outperform that within five years isn't bold. It's incurious about why the thing it's trying to replace actually works.

This is the pattern with the whole thread. The predictions that land are the ones grounded in something he clearly knows from experience — agency economics, audience monetisation, SaaS fragility. The ones that don't are the ones where he's pattern-matching from the outside, mistaking the absence of his own knowledge about a domain for the absence of complexity in that domain.

"I keep meeting people who are 'waiting for things to settle down' before they start building," he writes. "Things are not settling down." He's right about that. But the implied alternative — build everything, right now, as fast as possible — is how you get a thousand landing pages and no businesses. The window isn't closing on building. It's closing on building things that don't matter. Knowing the difference requires the one thing a coffee-fuelled prediction thread can't give you: the patience to actually understand the problem before you ship the solution.