The AI skill that got people hired in January is the one employers stopped asking for by summer. If you run a small team, the question is not which skill to teach next. It is how to build people who keep replacing their own.

Key takeaways

• The half-life of a specific AI skill is now under a year, so a curriculum built around tools is stale before it is finished.

• Most employees know what they need to learn and most say their employer is not helping, and that gap shows up as people who stay but stop caring.

• Small teams have an advantage here that large companies cannot buy: fast feedback, real problems, and the chance to learn in the open.

The Half-Life Problem

Workday published its global workforce report last week, and one number in it deserves more attention than it got. Demand in job postings for basic AI skills, the simple prompting that was the hot credential a year ago, peaked in January and then fell 25 percent. Over roughly the same period, demand for hands-on AI skills, people who can build tools and automate workflows, rose 51 percent.

Think about what that means for anyone who built a training plan in 2025. The thing they taught is already worth less. The thing they did not teach is what employers want. The cycle took about nine months.

This is not a one-time correction. Prompting was a skill because the models were awkward. The models got less awkward, and the skill dissolved. Building with AI is a skill today because the tools are new and rough. Those will smooth out too.

For a large company with a learning department, this is a planning problem. For a small business where the owner is the learning department, it is closer to a trap. You cannot afford to teach the wrong thing twice.

Staying Isn't Buying In

The second finding in the report worries me more, because it is about people rather than tools.

Of six thousand employees surveyed, 79 percent said they know which skills they need next. Only 66 percent said their employer helps them develop those skills. That thirteen-point gap has consequences. Internal moves fell at more than half of employers. Roughly half of employees did not try to move internally last year, and a quarter of those saw nothing worth moving toward. And employees who plan to stay are less likely to recommend their employer than those planning to leave.

Workday's phrase for this was that staying is not the same as buying in. I would put it more bluntly. A lot of people have decided their employer will not invest in their next skill, and they have responded by staying put and giving less. That is not loyalty. It is a rational response to a company that keeps telling people to reskill while cutting the paths that would let them.

McKinsey added the scale the same week, projecting that around eleven million American workers will need to change occupations by the middle of the next decade, and that only one in seven has a direct path to a job that pays as well. Its framing: the problem is mobility, not scarcity.

You will not solve the national mobility problem. But you are the entire mobility problem for the handful of people who work for you. Whether they grow or stall this year is mostly up to you.

Build the Builders

So what do you teach, when the specific skill keeps expiring? You stop teaching the skill and start building the capacity to acquire skills. Here is what that looks like inside a small team, because this is where small businesses have a real edge.

A large company teaches AI through a course. It picks a curriculum, licenses a platform, schedules sessions, and measures completion. By the time the course is approved, the tools have moved. A small company can do what a large one cannot: let people learn by solving the actual problem in front of them, with real stakes and same-day feedback.

The pattern we use at Stellant has three parts. Every person owns at least one workflow they are expected to improve with whatever tools exist, and to rebuild when better tools arrive. Nobody is handed a finished system and told to operate it. Learning happens in the open: when someone finds a better way, they show the team that week, and the showing is the teaching. And the measure of learning is not a certificate. It is whether a client or customer noticed something got better.

That is not a training program. It is a culture where the half-life of any skill does not matter, because the team is built to keep replacing them.

I spend much of my week inside research on how people learn with AI, and the clearest pattern so far is that people learn more when they are building something that matters to them than when they are completing a module about it. Small businesses have an unfair supply of things that matter. Use it.

The Learning Loop Audit

Here is a way to check whether your team is building capacity or consuming training. Answer these for each person, including yourself.

What is one workflow this person rebuilt in the past ninety days using a tool that did not exist a year ago? If nothing, they are operating, not learning, and it is not their fault. Nobody has handed them a problem and the room to solve it.

When did this person last show the team something they figured out? If you cannot remember, the learning is private or absent, and private learning does not compound.

Does this person know what skill they want next, and have you asked? Four in five employees know. Most have not been asked. Asking is the cheapest retention investment you have.

What would this person say if a friend asked whether to work here? If you are not sure, you may have the Workday problem: someone who plans to stay and would not recommend it.

What has your own learning looked like in ninety days? Teams learn the way their leaders learn. If you have not rebuilt one of your own workflows, nobody below you will feel permission to.

From Insight to Action

1. Assign every person one workflow to own and improve, with the clear expectation to rebuild it as tools change.

2. Hold a short, regular session where one person shows something they figured out. Keep it under twenty minutes and protect it.

3. Ask each person what skill they want next, this month, in a real conversation. Then do one concrete thing to help.

4. Stop measuring learning by completion. Measure it by what a customer noticed.

5. Rebuild one of your own workflows in the open, and tell the team what you got wrong. Permission to learn flows downhill.

6. Treat the people who plan to stay as carefully as those who might leave. Quiet staying is the more expensive problem, and the one nobody flags.

A company that teaches the current skill defaults to retraining. A company that builds people who keep learning defaults to growth.

Sources

Workday: "Global Workforce Report: AI Is Rewriting Jobs More Than It's Cutting Them," October 5, 2026

Fortune, reporting on the McKinsey Global Institute: "AI will create more jobs than it kills — after destroying 11 million," October 3, 2026