When Everyone Uses AI, How Do You Judge a Developer?

Source: When every programmer uses AI, how do we judge ability? (V2EX · Career)

Core takeaway

Ubiquitous tools do not flatten skill. The same AI in different hands produces very different outcomes.

Shift evaluation from “can they type code” to four signals:

1. Direction — how they frame and split the work
2. Course-correction — can they pull the model back when it drifts
3. Architecture & domain judgment — what to pick, what to refuse, how to cover edge cases
4. Maintainable delivery — not a pile of vibe-coded debt

What the OP asked

If everyone can use AI:

• Do interviews still grill textbook Q&A?
• Does pedigree matter more?
• Do likable, articulate people sail through more easily?

Under the hiring questions sits a sharper one: once coding speed stops being the yardstick, what replaces it?

Community consensus: same weapons, different fights

Recurring analogies:

• A strong eye technique is still only as strong as its wielder
• Everyone owning an AK does not make battles draw
• Same car, different drivers
• Calculators did not retire math fundamentals — “Google-oriented coding” becoming “AI-oriented coding” follows the same logic

Portable line: without human direction, AI digs deeper into the wrong hole.

One reported case: a strong engineer with a mid-tier model beat a rigid thinker with a top model. Expensive tokens do not patch weak judgment.

What rises, what falls

Rising weight: task understanding, architecture and business depth, exception coverage, communication, self-drive and learning speed, effective output per token.

Falling but not gone: raw typing speed, textbook trivia as the only filter. Weak vibing still shows up as unmaintainable, counter-intuitive systems — that is a signal.

Interview practice splits: keep fundamentals; keep ~80% and add change-adaptivity; take-home real features in days; or stop interviewing “programmers” and interview successor roles. Cold water: many firms still run leetcode + trivia next week.

Boss lens is blunt: story less, shipped value more.

Five checks for hiring or self-audit

1. Fuzzy-brief decomposition — direction and boundaries, not keystroke rate.
2. Drift catch — plant a bad AI path; see if they notice.
3. Maintainability — handover cost, failure modes, rollback — not just “it runs”.
4. Design why — algorithms and systems still fine; demand rationale.
5. Evidence of amplification — real results scaled by AI, not a “I use Cursor” line.

Job titles may change. Judgment and delivery quality will not. Before worrying how others get scored, ask which side of these five checks you sit on.

Related

• AI-era programmer survival guide
• AI economy and human jobs
• 35+ programmer crisis