Almeida’s bet is Jevons paradox, not a cheaper chatbot
In a run of replies three days after launch, the TypeSafe founder argued that Jev only works if cheap decisions create more software, not if they nibble LLM spend.
By September 18 the launch video was still circulating, and Almeida was answering the obvious business question: if you charge two orders of magnitude less, how do you have a company?
His reply is the Jevons thesis with a spreadsheet attached. TypeSafe, he said, does have a positive margin. But if AI consumption stays where it is, Jev captures very little. In his framing, every $100 taken from an LLM job is about $1 for TypeSafe, and less than half of LLM traffic even fits the interface. The upside is not a price war for chat. It is new loops that were never worth an LLM call: background checks inside software, run constantly, because they are cheap and typed.
That is also why he keeps saying he left OpenAI. Chat, in this telling, is an overfit interface for intelligence — great for humans, a bottleneck for machines. A follow-up post cheered the idea of making AI background technology so that software can be “fun and powerful” again instead of a transcript.
He also sketched a systems idea that is larger than Jev-the-API. Simple binary filters over a linear history still add up. If history were labeled as a nested tree of tasks, a model like Jev could search context in log(n) instead of rereading the log. Then you might include more agent history, more parallel workers, more subagents — because looking it up is no longer a full scan.
These are founder comments, not a roadmap with dates. They do explain the naming. Jevons watched 19th-century engines get more efficient and coal use go up. TypeSafe is trying to make a decision cheap enough that programs start making millions of them. If that fails, the company has priced itself into a niche. If it works, chat is no longer the unit of AI.