From authors to editors
When agents draft code, the engineer's value moves to shaping intent before, and judging output after. Promotion ladders, interview loops, and standups still measure 2019 work.
“If your interview loop is whiteboard-and-leetcode, you're hiring for a job that no longer exists at the senior level.”
Strategic implications
- 1Rewrite the senior engineer rubric around system design, code review, and skill authorship — not implementation speed.
- 2Replace at least one coding round with a review-an-AI-PR round. Calibrate on judgement, not typing.
- 3Standups should track unblocked decisions, not closed tickets.
Audit your last 10 promotions. How many were for review judgement vs. raw output? Adjust the rubric for the next cycle.
Personal opinion, not analysis. Dated above; revisit if the conditions change.
The work hasn't changed in our heads yet
Walk into a typical 2026 engineering org and the work appears normal: sprints, standups, PRs, code reviews. But look closer at how time is spent and the shift becomes obvious. Senior engineers spend more time framing problems for agents and reviewing agent-drafted PRs than they do typing code. The cognitive work has migrated from authorship to editorship.
The org's measurement of work hasn't migrated with it.
Where the friction shows up
Three places break first:
Promotion criteria
If your senior-engineer rubric still has "ships features quickly" or "writes clean code" as a top-line bullet, you're calibrating against authorship. Rewrite around: framing ambiguous problems, reviewing complex agent output, mentoring on judgement, designing systems that compose well.
Interview loops
A four-round whiteboard loop optimizes for raw implementation. That selects against the people who are most valuable in the new world — those who push back on a solution, ask the right questions, spot the edge case the agent missed. Replace at least one round with a "review this AI PR" exercise. Score on what they catch and what they ask, not what they would have written from scratch.
Standups
"What did you finish yesterday?" is a 2019 question. The answer is now either "I closed three agent-driven PRs" (which says nothing about quality) or "I'm halfway through reviewing a 30-hour run" (which doesn't fit the closed/not-closed frame).
A more useful standup question: what decision are you blocked on? That surfaces the actual bottleneck — judgement under uncertainty — rather than output volume.
Counter-argument
You can over-correct. Reading too much into "AI does the typing now" leads some orgs to fire IC engineers en masse and discover they no longer have anyone with the muscle memory to debug a hairy production issue at 3 AM. Editorship doesn't replace deep technical fluency — it requires it.
The point isn't to stop hiring engineers. It's to stop measuring them as if they're typists.