Lumen Autonomy

Built the self-improvement loop: the system can propose tweaks to its own settings, test them in a sealed sandbox, and — only if they clearly win and break no safety checks — hand them to a human to approve. Nothing self-promotes; everything is reversible.

Fed the assembly line real work, switched the live workers onto it, gave the main worker memory — then found seven in ten results dying and traced it to the checker being starved, not the work being bad.

Split the single worker into a three-stage line — plan it, write it, check it — and designed, on paper only, a careful way for the system to improve itself.

Built a flight-recorder-style safety net so the system survives a crash without coming back broken.

Gave the system long-term memory, plus a way to decide — per question — where to look for the answer.

Found the last untracked activity (not where we assumed), fixed it, and proved the new task flow works end to end.

Made every action the system takes visible and traceable. You can't trust what you can't see.

Decided what 'runs itself' actually means, then compressed an eight-week plan into a five-day sprint.