Submersible AI

Submersible AI is an agent experience paradigm in which the agent does not sit on the surface of the user’s tools, but operates at the user’s depth — continuously, inside the user’s real environment and workflow.

The agent sees what you see — your screens, your rooms, your physical obstructions — and you see what it sees: its reasoning state, its evidence, and its options, shown as a HUD, not buried in a chat log. Before you step into a task, it runs the task ahead in simulation and flags the traps. And it is vessel-independent: the same persistent agent, with the same memory and relationship — software today, hardware someday. The agent persists; the interfaces just keep getting less in the way.

In Michael’s own words, on the day he coined it: “Total submersible experience for AI — the agent can see what we see and even run simulations to see the obstructions we see / feel / experience in real time, and vice versa.”

On October 26, 2026, Michael and his AI operating partner spent roughly two hours getting through five separate sign-in locks just to publish one web page. The agent was blind the whole time. That is not a complaint about the agent — it is the design of nearly every AI tool in use today: the agent sits on the surface, guessing.

Every friction in that session was caused by an agent working at the surface. Submersible AI is the name for the alternative — named the same day, from first-hand evidence, and published for the record.

1. Lock 1

   Password — Google sign-in.

2. Lock 2

   Text code — SMS two-step verification.

3. Lock 3

   Email code — an emailed two-factor code.

4. Lock 4

   Device verification — GitHub device verification.

5. Lock 5

   Authenticator code — Vercel TOTP, relayed by hand against 30-second expirations, while screens were read second-hand, frame by frame.

6. Same day

   The first sketch — an interactive incident timeline rendering evidence as a holographic orb constellation: rooms of sourced fact you step into, instead of a page of text.