FIELD NOTE / LINKEDIN
Do not confuse less typing with less engineering.
The complete written thought and the evidence behind it. The video edition will follow its public release.
The written argument is here.
This approved LinkedIn edition is on the journal now. Its video player and original platform link will appear after each public release is verified.
Do not confuse less typing with less engineering.
Video caption
Do not confuse less typing with less engineering.
An elegant local patch may be unreleasable.
Some answers belong to business and service owners.
My rule: The person needs both code knowledge and judgment.
#EricFieldNotes
Full written post / accessibility read
I expect agents to write more of the code. That makes the engineer's causal model more important. A clean diff is one artifact. The system still has consumers, queues, identities, failure modes, product promises and a deployment sequence that the diff cannot settle on its own.
Show a compact fictional event migration with producer tests green. Ask the candidate to trace old-event replay, consumer compatibility, retry semantics, monitoring and rollback. A short code-reading excerpt can check whether they understand which assumption each test covers. Process language alone should earn no engineering pass; an unpaid build is unnecessary.
An agent can draft a compatibility branch. It cannot infer how long old events must remain valid or who accepts a customer-impacting cutover. The engineer should surface those choices, propose a discriminating replay test and keep the patch isolated until the right owners agree.
Require architecture, data-flow, security and operations fundamentals alongside attention and communication. Do this because an engineer who understands how the product works can direct agents to optimize or fix it after a context reset. Someone who only recognizes a convincing demo inherits a fragile dependency.
#EricFieldNotes
Four-beat scene transcript
1. Do not confuse less typing with less engineering.
I expect agents to write more of the code. That makes the engineer's causal model more important. A clean diff is one artifact. The system still has consumers, queues, identities, failure modes, product promises and a deployment sequence that the diff cannot settle on its own.
Visual: More plausible output increases review leverage.
2. Test fundamentals at real boundaries.
Show a compact fictional event migration with producer tests green. Ask the candidate to trace old-event replay, consumer compatibility, retry semantics, monitoring and rollback. A short code-reading excerpt can check whether they understand which assumption each test covers. Process language alone should earn no engineering pass; an unpaid build is unnecessary.
Visual: An elegant local patch may be unreleasable.
3. Separate implementation from policy.
An agent can draft a compatibility branch. It cannot infer how long old events must remain valid or who accepts a customer-impacting cutover. The engineer should surface those choices, propose a discriminating replay test and keep the patch isolated until the right owners agree.
Visual: Some answers belong to business and service owners.
4. Hire for the capacity to repair and extend.
Require architecture, data-flow, security and operations fundamentals alongside attention and communication. Do this because an engineer who understands how the product works can direct agents to optimize or fix it after a context reset. Someone who only recognizes a convincing demo inherits a fragile dependency.
Visual: The person needs both code knowledge and judgment.
Research and claim limits
- U.S. OPM: Work Samples and Simulations (S151)
- Google Research, Towards AI as a Collaborative Partner (S155)
- Vella and Blincoe, longitudinal AI coding-assistant study (S154)
The examples identified as illustrative or simulated are design probes, not reported incidents. Vendor specifications do not establish workload performance.