OpenAI Can't Hit Revenue Targets. Blame Agent Economics.

OpenAI missed key revenue and user targets on its path to an IPO, the Wall Street Journal reports. Converting curiosity into paying customers? Harder than it looks.

The agent economy explains a lot. We've moved from chatbots to thinking models to agents that actually do things. The problem? Agents burn through tokens at a rate Hacker News commenters call "truly insane." Each reasoning step costs money. Subagent calls and verification loops add up fast. Both OpenAI and its API customers feel the squeeze.

Competitors hit the same wall. Claude 3.5 Sonnet and Gemini 1.5 Pro push into agentic territory with tool use and multi-step reasoning. Same inference cost problems. Different logos.

Open-source frameworks like AutoGen, LangGraph, and CrewAI let developers build multi-agent systems without vendor lock-in. But they're still calling proprietary APIs for the actual thinking. The orchestration layer is open. The expensive part isn't.

The game has changed from model smarts to token efficiency. OpenAI's numbers confirm what builders already knew: the infrastructure for scalable agent workflows doesn't exist yet.