Compounding Engineering and Human Attention
Teams repeatedly correct agents for the same repository convention, testing requirement, architectural boundary, review preference, or failure mode. If those corrections remain inside individual conversations, the organization pays for the
A focused view of boundaries, contracts, state, authority, failure paths, and tradeoffs drawn from this chapter.
Reconstruct and defend this chapter’s architecture.
Reconstruct the architecture, name each boundary, and defend the tradeoffs.
Open the source exercise
Draw ten engineers repeatedly correcting the same repository mistake across three harnesses. Show correction capture, privacy and scope classification, clustering, deterministic-versus-skill decision, evaluation, promotion, canary, attention measurement, and rollback. Add one personal style preference that must not become an organizational rule.
4. Tradeoffs and alternatives
Capturing every correction creates surveillance, privacy, and noise. Capture only what serves a defined improvement purpose, minimize content, preserve consent and retention, and let users inspect or challenge derived preferences.
Personalization improves fit and can fragment team practice. Organizational standards improve consistency and can suppress legitimate variation. Keep scope explicit and allow local preferences only inside organizational policy and quality boundaries.
Reducing human time is not always the correct objective. High-consequence decisions deserve deliberate attention. Optimize away polling, repetitive repair, and context reconstruction—not accountability.
5. Current Mission Control Implementation
At study commit
d902fae,
Mission Control has human decision rights, risk-proportional approvals,
exception-first operator doctrine, decision packets, deterministic learning
signals, failure clusters, improvement candidates, datasets, experiments,
skills, context evaluations, canaries, and human promotion boundaries.
The studied evidence does not establish a production correction-harvesting pipeline, scoped Human Workflow Profiles, anti-pattern extraction, automatic suggestion of deterministic replacements, cross-team correction recurrence, or end-to-end attention accounting. Existing learning and telemetry mechanisms are suitable foundations but do not prove compounding engineering in operation.
Review this chapter.
Challenge a claim, boundary, missing failure mode, unclear term, or unsupported evidence statement.
- Claim
- Boundary
- Failure
- Evidence