GPT Operating Principle — Automation First / No Human Work When Machine Can Do It

Date: 2026-05-08
Recorded by: GPT-5.5 Thinking / Incomex Hội đồng AI
Source: User directive during P3D4C2U Option D execution window.

Principle

Những gì máy móc làm được thì không dùng con người.

All infrastructure and operation work should be designed as automation-first:

machine-verifiable
DOT/API/script-driven
repeatable
rollbackable
report-generating
no click UI
no manual checklist unless truly unavoidable

Governance implication

When a task includes a manual step, GPT/Opus/Agent must treat it as a possible system gap, not as a normal final state.

Manual steps are acceptable only when:

1. the system currently lacks the machine interface;
2. the gap is explicitly reported;
3. the object remains safe/draft/not-published if verification is incomplete;
4. the next pack proposes an automation or shared-module fix.

Apply to P3D4C2U

Current prompt allows route smoke to be CANNOT_VERIFY, in which case tbl_event_outbox must remain draft.

Under this principle, that outcome is not a final success. It means:

route_smoke=CANNOT_VERIFY
→ keep draft
→ identify missing automated route-smoke mechanism
→ next pack should automate route smoke or repair shared Table Module verification

Do not normalize human manual smoke as the long-term path.

System breakpoints to watch

For P3D4C2U and future packs, actively look for these gaps:

• DirectusTable route rendering cannot be smoke-tested automatically.
• Directus permission safety cannot be proven by intended-role token.
• Table Module field selector cannot be verified by code or request trace.
• DOT/API operation requires click-config in Directus UI.
• Publishing requires human visual confirmation when a machine-readable route check is feasible.
• Rollback relies on a human remembering what was changed.
• Report lacks machine-verifiable evidence.

Any of these should produce a follow-up shared-infrastructure repair pack, not a one-off workaround.

Rule for next directives

GPT should direct Opus/Agent to prefer:

automated smoke > manual smoke
API/DOT proof > visual inspection
field-level machine test > human trust
scripted rollback > hand rollback
shared module repair > local workaround

Boundaries

This principle does not eliminate legitimate human checkpoints for business decisions, approvals, or governance sign-off. It applies to operations that a system can verify or perform mechanically.

Humans decide policy. Machines execute and verify.