A hiring manager evaluating candidates in 2019 was, reasonably, looking for a fairly stable set of signals: years of relevant experience, a recognizable credential, deep specialization in a particular stack. Those signals made sense in a world where tools and best practices changed slowly enough that experience compounded predictably. That world doesn't describe the current one, and most interview loops are still built for it anyway.
The old signal set, and why it's weakening
Years of experience, credential pedigree, and deep specialization in one narrow stack all shared a hidden assumption: that the underlying tools and practices changed slowly enough for accumulated experience to keep compounding in value. That assumption held reasonably well for a long time. It holds much less well now. A candidate with deep, narrow expertise in a stack that's being reshaped by AI tooling every few months isn't automatically better positioned than someone with less tenure but faster adaptation, and pure years-of-experience as a filter increasingly screens for comfort with how things used to be done rather than capability with how things need to be done now.
The new signal set
Four signals now predict AI-era performance more reliably than the old set: tool fluency and adaptability (how fast someone actually picks up and productively uses new AI tooling, not just whether they've heard of it), judgment about when to trust versus verify AI-generated output (a genuinely new skill that didn't exist as a category before generative tools became load-bearing), raw speed of learning (since the specific tool or framework in front of someone today may not be the one in front of them in six months), and systems thinking across the AI stack (understanding how the model, the data, and the product layer interact, rather than being deep in only one layer and blind to the others).
| Old signal | Why it predicted less over time | New signal that replaces it |
|---|---|---|
| Years of experience | Compounding experience matters less when tools reshape every few months | Speed of learning and adaptation to new tooling |
| Credential/pedigree | A lagging, noisy proxy in a field this young and fast-moving | Evidence of real, recent shipped work |
| Deep specialization in one narrow stack | Depth in a stack being reshaped underneath you ages fast | Systems thinking across the model, data and product layers |
| Confidence in producing an answer | AI tools make producing an answer cheap; the answer alone proves little | Judgment about when to trust vs. independently verify AI output |
Why trust-vs-verify judgment deserves its own category
Generative AI tools can now produce plausible-looking output for almost any technical task, which means the bottleneck skill has shifted from producing an answer to knowing whether the answer is actually correct. A candidate who accepts AI-generated code, analysis, or content uncritically will ship confident-sounding mistakes at a much higher rate than one who has calibrated judgment about which categories of output need independent verification and which can be trusted. This didn't used to be a distinct interview-able skill, it's now one of the clearest differentiators between candidates who look similarly capable on the surface.
What this means for how interviews need to change
- Ask about a recent tool, framework, or model the candidate had to learn quickly, and probe how they actually approached ramping up, not just whether they eventually did.
- Include a task where some of the input or a draft answer is AI-generated, and see whether the candidate catches what needs correcting versus accepting it wholesale.
- Weight recent, verifiable shipped work more heavily than years of tenure or the recognizability of past employers.
- Ask candidates to reason across layers, model choice, data quality, product tradeoffs, rather than testing only the layer their resume specializes in.
- Treat 'I hadn't used this specific tool before, but here's how I'd figure it out' as a strong answer, not a gap, since that adaptability is now the more predictive signal.
What hasn't changed
None of this means experience or specialization stopped mattering, a candidate with real depth and a strong track record is still, all else equal, a safer bet than one without either. The claim is narrower and more specific: those signals alone no longer sort candidates as reliably as they used to, because the pace of change underneath most technical roles has outrun what tenure and narrow specialization were built to measure. Interview loops that haven't updated their weighting are still asking the old questions, and scoring candidates against the old bar, in a market that's already moved past it.
