Inside the AI credential gold rush, and what actually counts as proof
Too many job posts are demanding five or more years of production GenAI experience.
ChatGPT is three and a half years old.
Nobody on earth qualifies. So the roles go to whoever claims it loudest.
That one absurdity tells you where the AI talent market is right now. I’ve spent the past year on both sides of it. I review resumes for AI-adjacent roles. I’ve sat in one of the viral masterclasses being advertised to all of us. I’ve read the requisitions piling up on job boards. A credential economy is forming around AI, and it’s forming faster than any I’ve seen.
What I’m seeing in the resume pile
I’ve reviewed a ton of resumes over the last year, between my own hiring and the ones colleagues send me for a second opinion. The pattern I keep seeing is not a character flaw in the candidates. It’s the credential economy leaving fingerprints. Resumes stacked with recent AI certificates, because that’s what the courses promised would open doors. Bullets that describe using AI but never attach a number, because nobody said the measured result was the point. Earnest people doing exactly what the market told them to do.
For the record, I hold a stack of certificates myself. Certificates were never the problem. A cert beside a shipped system and a measured result is a credential. A cert standing alone is a promise the rest of the resume has to keep. Titles are cheap right now. Evidence isn’t.
Here’s what I almost never see, and what makes me sit up when I do: a candidate who writes “built an internal tool, adoption stalled at 20 percent, here’s what I changed and here’s where it landed.” That person gets a call every time. Failure plus measurement is the most credible thing you can put on a resume, and almost nobody writes it, because the credential economy has taught people that admitting failure is disqualifying. It’s the opposite.
What I found when I sat through one
I didn’t study this pattern from a distance. I signed up for one of these myself: a one-day AI masterclass from one of the fast-growing AI education brands whose ads had been following me around for months. Free entry, supposedly a limited-time waiver of the regular fee. I went in curious, and a little defensive. I lead an AI transformation for a living, and I wanted to see exactly what my colleagues and community were being sold.
Here’s my honest scorecard from that day. The actual instruction would have fit in a lunch break, and most of it is available free from the model providers themselves. The rest of the day was architecture: the founder’s origin story, income promises, a vision of the life you could be living, and a pitch that escalated as the hours passed, ending in a tiered offer with a deadline attached. There was a certificate, of course. Everyone gets the certificate. Mine went straight in the trash.
Since then I’ve read the landing pages and sat through the ad funnels of plenty more. They share a structure, and it isn’t a curriculum. It’s a sales funnel. A free masterclass that exists to sell the paid course. A paid course that exists to sell the “advanced” tier and the coaching program above it. Urgency everywhere: countdown timers, limited seats, prices that expire when the Zoom call ends. And the copy on more than a few of those landing pages carries the unmistakable rhythm of AI-generated text that nobody edited. Marketing generated in an afternoon, selling skills the seller may not have, to people who can’t yet tell the difference. That last part is not an accident. That’s the addressable market.
The pattern is older than AI
None of this is new. A discipline gets hot, demand for practitioners outruns the supply of people who’ve done the work, and a credential economy forms to close the gap. For a while the credentials are a decent proxy, because the early ones are built by genuine practitioners. Then the proxy becomes the product. Course creators optimize for enrollment instead of outcomes. Titles get claimed instead of earned. Within a few years the market can’t tell practitioners from performers, and everyone gets repriced together, including the people who were genuinely good.
You’ve seen this cycle before. It used to take a decade to play out.
With AI it’s happening in quarters.
The fix is not gatekeeping
The tempting response is to pull up the ladder. Demand credentials, restrict titles, form guilds. That instinct is wrong. It also doesn’t work. Gatekeeping is exactly how the last credential economy calcified.
The fix is changing what we accept as proof.
If you’re learning: skip the badge chase. Build something, break it, measure what happened, do it again. A weekend spent shipping a bad agent teaches more than forty hours of video, because the failure modes of real systems are the actual curriculum. The model that hallucinates your own data back at you. The prompt that works in testing and collapses on real users. The integration that costs ten times the estimate. No course can compress that, because the learning is the friction.
When you do want structured material, take it from people with receipts. Andrew Ng’s DeepLearning.AI courses are still the standard for fundamentals. Dr. Jules White at Vanderbilt has some of the most practical prompting material I’ve used. Anthropic, OpenAI, Databricks, Google, and Microsoft all publish serious free courses on their own platforms. Notice something about that list: the organizations building the technology give the education away, because their incentive is capable users, not course revenue. Remember that before you pay $499 for a funnel.
If you’re hiring: stop asking for years that don’t exist. Years of experience was always a weak proxy. For GenAI it’s a broken one. It filters out honest experts and selects for confident fabricators. I’ve replaced it in my own screens with three questions.
What did you ship? What broke? What did you measure?
Real practitioners light up at these questions, because real work generates stories about failure and measurement. Performers deflect, to credentials, to name-drops, to vision. If the answer to “what did you ship” is a certificate, you have your answer.
It’s the same standard I hold my own work to. The numbers that matter in the AI transformation I lead are boring and specific: 87 percent developer adoption, 39 percent faster feature delivery, measured quarterly, on a dashboard everyone can see. Not badges. Adoption and cycle time.
The counterargument I take seriously
There is one reason this cycle might end differently, and it’s a fair one.
In previous gold rushes, the product was advice, and advice is nearly impossible to audit. A performer could survive for years because outcomes were diffuse and attribution was murky. With AI, the product is working software. Shipped systems either function or they don’t. Cost is measurable, adoption is measurable, failure shows up in weeks instead of years. Maybe that feedback loop is fast enough to burn off the grift before it hardens into a permanent credential economy.
I want that to be true. I’m not convinced yet. Many organizations are still learning how to evaluate AI work on its merits, because the technology moved faster than the evaluation practices around it. Until they do, performers can survive on demos. The race is between how fast organizations learn to evaluate and how fast the grift professionalizes.
Which signal will you be
The AI talent market will eventually develop reliable signals. Every technology market does. The open question is how much value gets destroyed in the meantime, and how many capable people get repriced next to the pretenders before the sorting arrives.
You can’t fix the market. You can decide which signal you are. Build things. Measure them. Talk about what broke. In a market drowning in claimed expertise, demonstrated work is the only currency that appreciates.
I’m not saying burn every certificate. Mine just didn’t do the job it promised, and building things did. That’s the whole case.
I go deeper on all of this in my upcoming book The AI Leap, releasing soon: how leaders tell real transformation from theater, and how organizations get good at evaluating AI work. If you want the full playbook, that’s where it lives.
One last ask. What’s the wildest AI course promise or job requirement you’ve seen? I’m collecting them. Find me on LinkedIn and send it over. No names needed. Just the claim.