Hiring managers keep telling me the same story right now. The résumés are cleaner than ever, the cover letters are flawless, the take-home assignment came back perfect, and the person on the video call can't explain any of it.
A new survey puts a number on that feeling. Western Governors University released its Workforce Decoded report on September 30. It surveyed 3,128 U.S. hiring professionals between June 24 and July 7, 2026, fielded by Centiment. The headline: 60% of employers say AI has made it harder to evaluate candidates' real skills.
That is a majority of the hiring professionals surveyed, telling you the instruments they rely on are getting harder to read.
WGU's respondents aren't alone. Dawn Fay, operational president of the staffing firm Robert Half, said in a March survey release: "AI has made it easier to generate applications, but it hasn't made it easier to identify the right talent."
The numbers worth sitting with
- Nearly one in five cite difficulty confirming whether they are interviewing a human or an AI as a top challenge.
- The share of employers who say they're still figuring out how to evaluate AI skills effectively doubled, from 16% in 2025 to 32% in 2026.
- Among employers who say AI has made skills harder to evaluate, 54% report that AI has reduced entry-level hiring at their organization. Among those who don't report that difficulty, it was 20%.
That last one is the one I would put on a slide. It's correlation, and the report does not prove the evaluation problem causes the cutback. But my read is straightforward. When you can't trust the signal, you stop taking risks on people without track records. The juniors pay for the noise.
WGU President Scott Pulsipher framed it this way in the release: "AI is changing more than work; it is changing how employers identify and evaluate talent."
Why this lands hardest in technical hiring
If you hire engineers, you were already living with this. A coding take-home is now a test of whether the candidate has a good model subscription. A polished GitHub is a polished prompt history. A résumé keyword-matched to your job description was probably keyword-matched by a tool.
None of that makes the candidates dishonest. Using AI well is part of the job now. The problem is that the artifacts we used to treat as proof of skill no longer prove anything, and most interview loops still lean on exactly those artifacts.
What I would change this quarter
Here is what I'd tell a CTO or VP of Engineering. This is my opinion from the recruiting desk, not a finding from the WGU survey.
Start by testing judgment instead of output. Hand the candidate AI-generated code with a subtle flaw and ask them to review it live. Anyone can produce a working answer. Fewer people can say why this one is wrong and what it will cost in production.
Then make them narrate the work. Pick something on their résumé and go three questions deep: what broke, what they tried first, what they'd do differently. A person who did the work answers fast and specifically. A person who read about the work gets vague by question two.
Get a human into the loop earlier, too. Nearly one in five employers in the WGU survey name confirming they're interviewing a human, not an AI, as a top challenge. A fifteen-minute screening call from someone who knows the domain filters better than three rounds of automated assessment. Yes, it costs senior time. It costs less than a bad hire.
And don't stop hiring juniors because the signal got noisy. Pulsipher said that when employers can't trust the signals in front of them, opportunity narrows, "especially for people trying to get their first foothold in the workforce." I read that as a failure of the hiring process, and it's one you can fix without abandoning the entire entry-level pipeline. Paid trials, structured apprenticeships and project-based work give you a real look at someone before you commit. The report notes 38% of employers say internships, apprenticeships or project-based work are the most effective way for recent graduates to move from education into employment.
Last, write the req around evidence. "Describe a production issue you owned and how you resolved it" beats "5+ years Python" as a filter, and it's much harder to fake with a chatbot in thirty seconds.
Build the proof yourself
Pulsipher's answer: "Higher education and employers have to build a better proof system: credentials that carry meaning, assessments that verify what a person can do, and real-world experiences that show how they think, adapt and use AI responsibly." I agree with the direction. Waiting for universities and credential bodies to build it is a slow plan, though.
You can build your own version of it by next month. Fewer artifacts, more live work, and a screening conversation led by someone who knows what good looks like.
AI didn't break hiring. It exposed how much of it was running on proxies. The take-home will keep coming back perfect. Ask them to explain it.
Sources: WGU Workforce Decoded report release via GlobeNewswire, September 30, 2026; Robert Half survey release, March 10, 2026.