Emanuele Melis is a lead field engineer at AI One, which builds Context One: a context control layer between enterprise systems that gives LLMs and agents accurate, governed context. The idea is that big companies can automate complex workflows without first running a large data transformation program. He joined us in Episode 46 (June 15, 2026), the week Anthropic released Fable.
AI One is a New York City-based team with a background in high-performance regulated trading and data platforms in New York and Europe. It started about two years before the episode. Emanuele's team is the one that installs Context One at a customer, onboards them and builds their first use cases on top of it.
Emanuele's example was customer onboarding at a large enterprise. A questionnaire moves from team A to team B to team C. Team D finds an error it can't fix, so it goes back to team A, then to sales, then to the customer, and something simple takes a month. AI One's answer is to let agents handle as much of that chain as possible, with the human coming in only at the last mile.
The company targets what he called high complexity, low judgment work. Older systems handled edge cases by stacking rules until a process had 150,000 of them; he said agents deal with the nuance and edge cases much faster than a set of rules.
You start every agentive workflow with the autonomy slider set to zero. So the agent really has no autonomy.
— Emanuele Melis, Episode 46
Emanuele opened at 02:17 with a game, not AI One. As a kid in the 1990s he played a puzzle game where you build contraptions, but he could no longer remember its name or find it. When Fable came out, it was the first thing he tried: he described the game and asked for it in a web UI. The result has you place objects like a trampoline to bounce a ball into a basket, and he got it working on air after a few tries. He gave different times for the build, from a couple of minutes to ten. Jordan was most impressed that the physics behaved realistically on the first go.
What's crazy to think about it is that a year ago, this would not have been possible, but today with with Fable, literally took five minutes to put this together.
— Emanuele Melis, Episode 46
In the same episode, Jordan shared his first 36 hours with Fable, after usually driving Codex as his daily tool. He found it slow and seemingly expensive, with confusing effort settings, but accurate, and said typing "100x the design" produced world-class design work. Compare the tools on our Claude Code vs OpenAI Codex page, and see AI in finance for more on agents in regulated work.
Emanuele Melis is a lead field engineer at AI One, a New York City-based team with a background in high-performance regulated trading and data platforms in New York and Europe. His team installs Context One for customers and builds their first use cases on it.
AI One builds Context One, a context control layer that sits between enterprise systems. LLMs and agents use it to get accurate, governed context, so enterprises can automate complex workflows without first running a large data transformation program.
Every agent workflow starts with autonomy at zero: the agent only drafts and researches, and a person takes the next action. As confidence and knowledge in Context One grow, the slider moves up, maybe to 30 or 40%, and Emanuele says it may never reach 100%.
AI One targets high-complexity, low-judgment work. Emanuele's favorite example is a regulated financial investigation where agents check transactions, documents and logs in parallel, turning about a week of work into a couple of hours.
