The chatbot era is ending. The agent era just cost its first $2B.

For two years, every serious AI founder deck opened with the same slide: "We're the ChatGPT of X." Legal, HR, accounting, customer support, creative — all "chat interfaces for domain experts." By Q1 2026, that slide is a tell. The money has moved.

The question stopped being "can you answer" and started being "can you act."

What's funding now is a different species. Not chat. Work. Three companies — one in enterprise procurement, one in legal contract review, one in embedded finance — crossed a collective $2B in valuation in ninety days. Their pitch decks share three sentences the chatbot generation never needed: "Runs unattended. Books calendar events. Makes the call."

Founders who still sell "conversational AI" are fighting the last war. Two signals matter now.

First: revenue per customer doubles when the interface disappears. A chatbot that answers 1,000 tickets a week costs less than a human. An agent that closes 200 tickets — end-to-end, including the refund — costs less than the chatbot. Buyers stopped asking for a tool; they started asking for an outcome.

Second: the winning companies quietly built their own evals, their own observability, their own human-in-the-loop harness. The moat is not the model. The moat is the operating discipline around the model.

What this means for a founder reading this at a coffee shop: if your product ships a chat interface as its primary surface, you have about four quarters to find your agentic hook. Verticalized, opinionated, and — critically — measured against work done, not words generated.

The cycle will keep moving. In 2027 the narrative will shift again, probably to "systems of systems." The operators who survive the next turn will be the ones who stopped trying to be the interface, and started selling the result.