We solved the "AI agent black box" problem with typed tasks

Most tools store plans in plain markdown that agents forget about. We built Tasklist to fix this by making each step a first-class object with strict state transitions (todo → in_progress → finished).

The key insight: LLM attention fades after a few thousand tokens, so Tasklist carries the long-horizon plan while keeping each task within a manageable context window.

Real-time visibility means you can watch agents work, stop them mid-execution, or redirect when things go sideways. Each task emits structured events that enable analytics and future automation.