FIELD NOTE / X
The engineer becomes the decision owner.
The complete written thought and the evidence behind it. The video edition will follow its public release.
The written argument is here.
This approved X edition is on the journal now. Its video player and original platform link will appear after each public release is verified.
The engineer becomes the decision owner.
Video caption
The engineer becomes the decision owner. The team re-prompts, retests and redeploys. Ship on observed behavior, not polished code volume. #EricFieldNotes
Full written post / accessibility read
When agents write more code, the engineer's scarce work is naming which behavior is acceptable. Consider a migration that could preserve legacy entitlements or normalize them. Both implementations can be clean. Only one matches the product and customer promise.
If nobody settles the legacy rule before generation, agents can build and test against different assumptions. After product finally answers, the team must trace every affected route and repeat validation. The cost is attention and flow disruption, not necessarily hand-recoding.
The engineer lists the two entitlement policies, the users each affects, migration behavior and rollback condition. Product names the chosen outcome. A protected old-record fixture distinguishes the branches. Agents then implement against that versioned contract.
Let agents write implementation and supporting tests. Keep the business fixture and outcome verdict under independent control. Do this because the engineer adds leverage by making the important decision testable before a hundred fast patches encode the wrong one.
#EricFieldNotes
Four-beat scene transcript
1. The engineer becomes the decision owner.
When agents write more code, the engineer's scarce work is naming which behavior is acceptable. Consider a migration that could preserve legacy entitlements or normalize them. Both implementations can be clean. Only one matches the product and customer promise.
Visual: Agents can write a patch without choosing the right tradeoff.
2. Late judgment makes fast code expensive.
If nobody settles the legacy rule before generation, agents can build and test against different assumptions. After product finally answers, the team must trace every affected route and repeat validation. The cost is attention and flow disruption, not necessarily hand-recoding.
Visual: The team re-prompts, retests and redeploys.
3. Write a decision packet first.
The engineer lists the two entitlement policies, the users each affects, migration behavior and rollback condition. Product names the chosen outcome. A protected old-record fixture distinguishes the branches. Agents then implement against that versioned contract.
Visual: Alternatives, owner, evidence and a discriminating probe.
4. Own the invariant; delegate the bulk.
Let agents write implementation and supporting tests. Keep the business fixture and outcome verdict under independent control. Do this because the engineer adds leverage by making the important decision testable before a hundred fast patches encode the wrong one.
Visual: Ship on observed behavior, not polished code volume.
Research and claim limits
- METR randomized trial (S20)
- Stack Overflow 2025 developer survey (S21)
- METR 2026 productivity experiment update (S93)
The examples identified as illustrative or simulated are design probes, not reported incidents. Vendor specifications do not establish workload performance.