Skills gaps that stall AI from pilot to production

A founder and CTO in advanced materials, an internationally recognized authority on his field, told me he has been an engineer for thirty-five years without an engineering degree. He started in a plastics shop, learned carbon fiber on the job, and ended up teaching seminars about graphene to rooms full of people who hold PhDs in organic chemistry. He does it, he said, by taking a lot of little blocks of information and building a story he can understand, then rebuilding those blocks for other people. He has trained more than fifty engineers this way. His view is simple: you will learn wherever you are, if you want to learn.

I keep that story close when teams tell me their AI pilot is stuck on a skills gap. Usually they mean they are missing a specific title, a machine-learning specialist or a data engineer. Sometimes that is true. More often the real gap is the one the CTO spent his career filling, the ability to learn a new thing quickly, and no job posting screens for it.

The gap is usually learning speed, not a credential

Around 38% of teams cite a lack of role-specific training as a barrier to AI adoption. That number is real, but it points somewhere more useful than “hire a specialist.” The people who move an AI pilot into production are rarely the ones who already knew how. They are the ones who could learn how in weeks, because the tools change faster than any curriculum keeps up with.

The CTO’s method is the whole point. He does not memorize a field. He breaks it into small blocks, builds understanding, and then teaches it forward. That is exactly the skill AI production rewards, because the model you use, the framework, the evaluation approach, all of it will be different in a year. A team of people who can decompose a new problem beats a team of people with the right titles and no appetite to learn.

He said the degree used to mean you could stick with a program and develop a line of reasoning. Now, he said, it is really about what you know and how you apply it. In AI work, what you knew last year is already half stale. How fast you learn the rest is the job.

What to actually assess

Can they learn in public? The people who move fast ask questions early and share half-formed understanding. The ones who hide until they are sure are slower than they look.
Do they decompose or do they wait? Give someone an unfamiliar problem. Watch whether they break it into blocks or stall until an expert appears. The first behavior scales, the second does not.
Have they taught something recently? Teaching is the proof you actually understand. The CTO teaches PhDs precisely because he built the blocks himself.
Are they attached to a tool or to a result? People attached to a specific stack fight the next change. People attached to the outcome adopt whatever gets there.

How we approach it at Density Labs

Our AI Readiness Assessment ($2,500) assesses the capability gap honestly, and it is rarely the one teams expect. We look at whether the people who will own this in production can learn the next version of it, not just run the current one. Sometimes the answer is a hire. Often it is that the right person is already there and simply has not been pointed at the problem. Knowing which one you face is worth far more than another job posting.

The credential tells you what someone learned once. Production is a bet on what they can learn next. Hire, and staff, for the second.