FIELD NOTE / LINKEDIN
Can a second person change the service?
The short film, the complete written thought, and the evidence behind it.
The LinkedIn conversation link will follow its public release.
Can a second person change the service?
Video caption
Can a second person change the service?
Rationale, boundaries, tests and rollback are explicit.
The test remains independent of the worker.
My rule: Time to an accepted safe change is the measure.
Narration uses Eric's authorized AI voice clone.
#EricFieldNotes
Full written post / accessibility read
An internal app is not truly maintainable if only the original prompt writer understands its policy shortcuts. The meaningful acceptance exercise is a bounded second change handled by another operator.
Provide the versioned decision log, state diagram, affected service boundaries, protected cases and incident trace format. Ask the operator to explain which outputs should change before running another agent session.
The new operator can delegate code generation, but the expected state table and regression harness live outside the editable worktree. The exercise ends at an authoritative readback and a named rollback condition.
Compare review effort, surprises and the ability to recover without the original session. Do this because architecture is useful when another person can extend the product without rediscovering what the business meant.
Narration uses Eric's authorized AI voice clone.
#EricFieldNotes
Four-beat scene transcript
1. Can a second person change the service?
An internal app is not truly maintainable if only the original prompt writer understands its policy shortcuts. The meaningful acceptance exercise is a bounded second change handled by another operator.
Visual: A codebase and tidy summary may still require its creator.
2. Give them a compact operating packet.
Provide the versioned decision log, state diagram, affected service boundaries, protected cases and incident trace format. Ask the operator to explain which outputs should change before running another agent session.
Visual: Rationale, boundaries, tests and rollback are explicit.
3. Let agents implement within that contract.
The new operator can delegate code generation, but the expected state table and regression harness live outside the editable worktree. The exercise ends at an authoritative readback and a named rollback condition.
Visual: The test remains independent of the worker.
4. Evaluate transfer, not theater.
Compare review effort, surprises and the ability to recover without the original session. Do this because architecture is useful when another person can extend the product without rediscovering what the business meant.
Visual: Time to an accepted safe change is the measure.
Research and claim limits
- DORA 2025 State of AI-assisted Software Development (S225)
- GitHub Copilot Agents application card (S226)
- NIST AI RMF Core (S227)
The examples identified as illustrative or simulated are design probes, not reported incidents. Vendor specifications do not establish workload performance.