The daily loop
Proactive collection, human prioritization, gated execution. How APD/M runs every day without waiting to be poked.
Proactive collection, human prioritization, gated execution. How APD/M runs every day without waiting to be poked.
The heart of APD/M is not a clever prompt. It is a loop that runs whether or not someone remembered to open the chat.
Product work dies in coordination tax: status hunting, stale boards, metrics nobody pulled, decisions that lived in a thread and then evaporated.
A daily loop reverses that:
1. Collect health and status with scripts first (service health, board, feedback, infra signals)
2. Cross-check against commits and real artifacts, not a ghost task list
3. Surface a short, human-readable picture before the working session
4. Decide (humans) what matters today
5. Execute with agents under review
6. Learn into the next day's collection
When this runs, standup stops being a roll call and starts being a decision meeting.
Every morning the product-squad system assembles the picture: collectors report health, board state, and feedback; agents synthesize against recent commits; humans prioritize; specialists pick up execution with review and a human ship gate.
That cadence is the operating system. Chat is the interface. The loop is the product.
Scripts collect. Models interpret. Humans prioritize. Agents execute with gates.
If you reverse that order, you pay twice: tokens for fiction, then human time for cleanup.
• Treating messages sent as loop success
• Letting collectors invent a pretty narrative on top of bad metrics
• Skipping human morning judgment and calling that autonomy
• Automating against a false board
Ground truth first. Throughput second. Trust is the constraint that makes speed real.
Reactive chat is how you start. A daily loop with trustworthy collectors is how a small team stays pointed at the right work while agents multiply capacity. That is APD/M management, not a demo script.