Skip to content

November 17, 2026

Can You Teach AI What Took You 20 Years to Learn?

The words the company brain on a blue background with a subtitle From memory to intelligence.

Imagine sitting down with a new employee on their first day.

They’re incredibly smart.

They’ve read every book about your industry.

They understand the terminology.

They know the regulations, the frameworks, the processes, and the best practices.

You ask them a complicated question.

They give you a textbook-perfect answer.

Then you say:

“Okay. Now let me tell you what actually happens.”

Anyone who’s spent enough time in business has probably had some version of that conversation.

There’s what you’re taught.

And there’s what you learn after doing the job for 20 years.

I’ve been thinking about whether AI is reaching a similar point.

Large language models have already absorbed enormous amounts of information.

They’re remarkably good at producing answers.

But professional work often involves situations where the answer isn’t obvious.

The customer is asking for something unreasonable.

The contract contains an unusual provision.

The project is technically on schedule, but something feels wrong.

The financial numbers are correct, yet the underlying assumptions don’t make sense.

An experienced professional knows how to navigate those situations.

So how do we teach AI to do the same?

One increasingly important approach is post-training.

The concept is straightforward, even if the technology behind it is complicated.

You start with a model that already has broad capabilities.

Then you improve its performance on particular types of tasks.

One method is supervised fine-tuning, where the model learns from examples of good work.

Another involves reinforcement learning, where the model receives feedback or rewards based on how well it performs.

The interesting part is determining what deserves a reward.

For simple tasks, that’s relatively easy.

Did the code pass the test?

Did the calculation produce the correct result?

Did the system follow the required instructions?

But what about evaluating a complicated legal argument?

Or deciding whether a customer relationship is worth pursuing?

Or assessing whether a project should be stopped before it consumes another $10 million?

Those decisions involve tradeoffs.

And that’s where experienced professionals become important.

A senior lawyer can explain why one argument is stronger than another.

An experienced physician can identify which clinical factors should influence a recommendation.

A seasoned executive can recognize when a technically reasonable decision is likely to create problems later.

Their expertise can help create examples, evaluation criteria, and feedback that improve AI performance.

Harvey’s research on post-training legal models offers a useful illustration.

Its work involves realistic legal assignments, expert-designed grading criteria, and reinforcement learning to improve how models perform complex legal tasks.

That’s a meaningful shift from simply giving an AI access to more legal documents.

But there’s a catch.

Twenty years of experience can’t be transferred in a weekend.

Much of professional judgment is difficult to explain.

Some lessons are specific to a company, industry, or situation.

And even experienced professionals disagree.

Building reliable AI requires more than collecting expert opinions. Those opinions need to be tested, structured, and evaluated against actual outcomes.

Still, I think this is one of the most promising directions in enterprise AI.

Because the goal isn’t to make every employee spend 20 years learning the same lessons.

It’s to help the next generation benefit from what the previous generation already figured out.

Imagine a new consultant who can draw on decades of implementation experience.

A junior engineer who can benefit from thousands of difficult troubleshooting situations.

A young attorney who can access the judgment behind years of complex legal work.

They’ll still need to develop their own expertise.

They’ll still make decisions.

They’ll still be accountable for their work.

But maybe they won’t have to start from zero.

And that might be one of the biggest opportunities AI creates.

Not eliminating the value of experience.

Making it available to far more people.

 

Looking to take your AI Agent from pilot to production? Check out Consulting Solutions full range of capabilities here and then contact us to request a consult.

Ready to see what integrated delivery actually looks like?

Tell us what you're working on. We'll bring the right people to the conversation — usually within a few days, not a few weeks.

Request a Conversation