Edward Levin

AIR is short for AI Resource — a direct analogy to HR, Human Resources.

AIR started as a way to fill a gap that many startups and small businesses run into. A business needs a whole range of skills to research, build, market, and sell a product, and it rarely has all of them on hand. AIR — under a different name at the time — was the answer: while you are still searching for the right hire or raising the funds to bring a specialist on, AIR can stand in for those roles as a cooperative AI teammate.

To make that work, I had to solve three problems. How can a prompt-based AI teammate be trusted? How do you keep work continuous from one session to the next? And how do you make it genuinely portable, with no vendor lock-in? Prompt-based AIR answers each of them, to an extent.

I approached trust through transparency. The AIR objects that print into the chat — structured JSON blocks — show exactly how AIR is executing every task, out in the open, so none of its reasoning stays hidden.

Continuity comes from the handoff card: a JSON block that captures where the project stands and what the next step should be. Because JSON travels everywhere, that card carries a project not just into a new session but onto an entirely different platform, so the work continues seamlessly even as the AI landscape keeps shifting.

That same portability frees AIR from any single model or platform. You boot a new session wherever you like and carry on from the card. The one honest caveat is that the output is always shaped by whichever model happens to be hosting it.

This is only the surface. I would suggest reading more about the methods and concepts behind AIR's workflow — or simply booting a session and seeing it for yourself.