Ever since I can remember, I've wanted to be an AI engineer. after Goodfellas.
SEVEN AGENTS. NO PROTOCOL. FILES ONLY.
ONE ADVERSARY. DIFFERENT MODEL. DIFFERENT OPINION.
EVERY WRITE IS A DECISION YOU CAN CITE.
The fleet uses files as source of truth. Not because files are elegant. Because files are inspectable when something breaks. A handoff can be read without booting the system. A missing file is a visible failure. A new agent enters the work by reading state that already exists.
That's the reason I chose folders over a clean orchestration layer. SDKs are faster when the happy path holds. Files are better when the question is: what happened, who decided it, and what changed.
Seven colour terminals do domain work. One adversary — a different model, deliberately — argues with public claims before they ship. For material choices I use overlap across models because agreement from the same blind spot is cheap.
Every write should be a decision you can cite. That's the standard this site is being forced through. Claim. File. Timestamp. Reason.
Files as coordination are slower than in-process message passing. That's part of the trade. I chose inspectability over elegance because the failure mode matters more to me than the happy path. The fleet benchmark is set up to measure where files help and what they cost. Until it runs, I don't publish a win claim.
Cross-terminal dispatch has a fragile layer. Keyboard automation works until the operating system changes a permission or a window isn't where the script expects. I monitor the flows that need to stay up. The fragility is real.
The adversary can miss things the fleet also misses. Different model, different training, but not orthogonal. When the adversary approves something, that doesn't prove it's right. It proves it survived a second, different check. Better than one. Not proof.