Plan for Who Learns, Not Just What AI Does

Planning 2027 with AI in the Room · November 2026

It's November, and in most organizations I work with, next year's budget closes in a few weeks. This year the plans have a new kind of line in them: AI. A pilot for customer service. A tool for drafting reports. An agent for reconciliations.

Each of those lines comes with a business case, and almost every business case scores the same thing: value. Hours saved, cycle time cut, cost avoided. That's a fair question to ask. It's just not the only one.

The cost that doesn't show up in the plan

Routine work has always done two jobs at once. It produced output, and it trained the people doing it. The new analyst who reconciled hundreds of ordinary exceptions learned which ones matter. The junior consultant who wrote a dozen first drafts learned what good looks like, because someone more experienced marked them up.

When AI takes over that work, the first job gets done faster. The second job quietly stops. Nobody decides to stop it. It just isn't on anyone's plan.

The cost doesn't arrive in 2027. It arrives two or three years later, when you look for people ready to take on harder calls and find that the practice which would have made them ready never happened.

A second score for every AI use case

So here's the one change I'd make to a 2027 plan: score every AI use case twice.

The first score is value, as you already do. The second is learning at stake:

Learning at stake is the second score for each AI use case, from 1 to 3: who learns on this work today, and what replaces that learning if AI takes it. It is scored alongside value.

A simple way to set it:

• 1: little at stake. Nobody is really learning on this work today, or people already learn the same thing elsewhere.

• 2: some at stake. People do learn on it, and you can name how they'll keep learning once AI does most of it.

• 3: a lot at stake. This work is how newer people build judgment (the ability to make good calls), and you don't yet know what replaces it.

A use case with high value and a 3 for learning at stake isn't a no. It's a yes with a second line item: what you'll put in place so people keep growing on the work. That might be having newer people attempt the task first and then compare with a reference answer, a short growth target for the people affected, or a peer circle. It doesn't have to be big. It has to be on the plan, with an owner.

Three questions for your planning meeting

If you're in a planning meeting this month, these three questions take about ten minutes per use case:

1. Who learns on this work today? Name the roles, not "the team".

2. If AI does most of it, where will they learn it instead? If the answer is "on the job", ask which job.

3. Who owns that, and is it in the budget? A growth practice added to someone's job without taking anything away usually doesn't survive the first busy quarter.

You won't have perfect answers. You'll have better ones than a plan that never asked.

Why I'm writing this series

I've been planning my own 2027 this month, and the most useful decisions were the ones about what not to fill: room left on purpose, checkpoints set before the year starts, and clear rules for when to change course. The same habits help any leader planning around AI. This series takes them one at a time, every other week, through February.

For leaders: Before you sign off on an AI line item for 2027, ask for its learning-at-stake score. If nobody can tell you who learns on that work today, that is your first finding.

Which AI use case in your 2027 plan would score a 3, and who owns what replaces it?

Next in the series: Leave Slack on Purpose (early December)

Back to the series → Planning 2027 with AI in the Room · Get in touch →