Agent orchestration · Running daily
The Fleet
Agents that delegate to agents. Research, build, and review lanes with defined roles, reporting back to a phone.
Why I built it
One assistant doing everything gets sloppy. Splitting the work into lanes with clear briefs, and making one agent responsible for checking the others, produced better results and fewer surprises.
What it does
- Roles, not one chatbot
- Each agent has a lane and a brief. The orchestrator plans and verifies; it doesn't do the work itself.
- Persistent memory
- Decisions, corrections, and daily logs are written to disk, so a new session starts where the last one ended.
- Reports to a phone
- When a task completes or needs input, a message lands on my phone. No watching a terminal.
- Verification is mandatory
- Nothing is called done until the output is checked. Screenshots, greps, tests — whatever the task needs.
the fleet2026
- ›orchestrator · plan · 7 phases
- ›brief → frontend · hero section
- ›brief → qa · scroll capture 1440/390
- ›verify · 42 screenshots · 2 seams
- ›fix · seam at 3600px
- ›notify · phone · ready for review
- Claude Code
- Custom agent roster
- File-based memory
- Messaging bridge
What it taught me
This is the same pattern I set up for clients at a smaller scale: one clear brief, one place the work lands, one check before anything ships.