FIELD NOTE / TIKTOK
Test what the build brief omitted.
The short film, the complete written thought, and the evidence behind it.
The TikTok conversation link will follow its public release.
Test what the build brief omitted.
Video caption
Test what the build brief omitted. Guessing active would create an unauthorized state. Release only after the exception is owned. Narration uses Eric's authorized AI voice clone. #EricFieldNotes
Full written post / accessibility read
Want a harder test for an agent-built approval app? Keep one real kind of historical exception out of the build brief. The builders see the clean path; the acceptance owner holds the exception.
In this proposed case, the legacy Boolean says approved but the approver field is missing. That does not prove the customer should receive access. The safe expected outcome may be unknown.
Before revealing the owner's answer, ask the implementation to predict what it will store, what the audit log will say and what the customer will see. Run the case in an isolated fixture.
Compare all three observations with the protected owner label. Do this because the strongest test is the case missing from the prompt, and uncertainty should not silently become permission.
Narration uses Eric's authorized AI voice clone.
#EricFieldNotes
Four-beat scene transcript
1. Test what the build brief omitted.
Want a harder test for an agent-built approval app? Keep one real kind of historical exception out of the build brief. The builders see the clean path; the acceptance owner holds the exception.
Visual: The unknown case matters more than another clean demo.
2. The held-out record has no clear approver.
In this proposed case, the legacy Boolean says approved but the approver field is missing. That does not prove the customer should receive access. The safe expected outcome may be unknown.
Visual: Guessing active would create an unauthorized state.
3. Predict three observations first.
Before revealing the owner's answer, ask the implementation to predict what it will store, what the audit log will say and what the customer will see. Run the case in an isolated fixture.
Visual: Persisted entitlement, audit entry and visible status.
4. Treat unknown as a designed answer.
Compare all three observations with the protected owner label. Do this because the strongest test is the case missing from the prompt, and uncertainty should not silently become permission.
Visual: Release only after the exception is owned.
Research and claim limits
- 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.