Why Tech Hiring Stopped Trusting Skills Lists

Since January 2025, US tech job postings have been asking for more years of experience, and fewer specifically listed skills. The shift, measured by Revelio Labs and relayed by a16z, looks modest on paper: roughly 5 to 10% in both directions. But it's been consistent for a year and a half, not a one-off statistical blip.

a16z itself stays careful about the cause. The existing tech workforce is aging in general, and that can be explained by several things that have nothing to do with AI: people staying longer in their roles, remote work widening the pool of candidates a recruiter can reach, a general premium on seniority that predates any of this. On that broad point, the newsletter says so directly: the evidence tying it to AI is weak.

Where the signal gets sharp is in looking at which specific roles carry the biggest premium on experience: data work, systems work, and Java sit clearly at the top. Those aren't random categories. They're the two or three areas AI has upended the most this year.

The problem of uncharted ground

The explanation comes down to one simple idea: on ground nobody has mapped yet, a listed skill doesn't mean much, because the reference point for what counts as good practice barely exists yet. The right way to structure a data pipeline to feed a model, or decide what to automate versus keep manual, changes faster than any standard can settle. Data wrangling in particular has changed shape in a very short window, and no one has a settled answer yet on the right way to do it.

When nobody quite knows yet what needs to be done, the skill itself hasn't been defined. What's useful in that gap is someone who can move without a playbook: diagnose a poorly defined problem, try something, get it wrong, correct course, without waiting for an official standard to tell them how.

What a resume verifies, and what it doesn't

A tech stack on a resume checks out in five minutes: either the person has used it or they haven't. The ability to navigate a problem that doesn't have a documented right answer yet doesn't check out the same way. It takes digging into a real example, not reading a line item. It's slower to evaluate, and that's exactly why the natural reflex, in hiring and in picking a technical partner alike, is to fall back on the verifiable list rather than the question that actually matters.

The labor market is starting, slowly, to correct that reflex: it's already paying a bit more for experience than for the checklist, precisely where the ground is shifting fastest.

Two honest limits worth keeping in mind

First, experience alone isn't a totem. A too-quick reading of this data would push toward hiring only seasoned seniors, which has its own cost: it shuts the door on people who, precisely because they don't have old habits to unlearn, sometimes adapt faster to unfamiliar ground. The real filter isn't years on the job, it's evidence of having already held a course without a ready-made framework, which can show up early in a career.

Second, and this is the most interesting part of the a16z data: this imbalance is likely temporary. As AI use around data and systems matures, new skills will eventually get defined and documented. Once that happens, the balance between experience and listed skills shifts again, probably back toward skills. The current premium on experience isn't a permanent truth about the value of experience in general. It's a symptom of ground that doesn't have a map yet.

What this changes, today

For judging a hire or a technical partner on ground that's still shifting, a skills checklist isn't wrong, it's just premature: the reference point it claims to verify doesn't really exist yet. Until it settles, the more valuable question stays the same: show me a decision you made without a playbook, and how you built it.