Should AI training come before AI tools? What to cover

The training foundation should come before a firm-wide license purchase, and it should be built on a handful of licenses and one real piece of work. The foundation is four working habits, and a team that lacks them will get little from any tool you buy.

That is a narrower claim than "train first," and the difference matters when you are deciding what to approve this quarter.

What happens when the licenses arrive first

Here is an illustration of the pattern, not a report on any one firm. Everyone gets a login on a Monday. A few people use it to tidy up emails. One analyst gets very good at it and does not tell anyone how. Three months later the invoice renews and the lease abstracts, the client letters and the monthly reports are produced the same way they were before.

The tool is rarely what went wrong. Nobody showed the team which parts of their own work to hand over, what the tool needs to be told, how to check what comes back, or who answers for the result.

The survey data suggests this order of events is the normal one. The Federal Reserve Bank of New York found in its November 2025 Survey of Consumer Expectations that 39 percent of employed respondents were using AI tools in their job or had done so in the past twelve months, while only 15.9 percent said their employer offers any AI training. A University of Konstanz survey of 1,105 employees in May 2026 found that only 11 percent of employees in small organizations report having received AI training, and that only 55 percent of AI users say the tool they use most was officially introduced by their employer.

What is a training foundation, in plain terms?

A training foundation is the set of habits a person needs before any AI tool is useful to them on client work. It is independent of the product. A person who has it can move from one assistant to another, or from a chat window to a built workflow, and stay productive. A person who lacks it will get weak results from all of them.

We think a course list is the wrong way to define it. A partner cannot look at a list of modules and tell whether the team can do the work. A partner can watch someone abstract a lease or draft an engagement letter with AI and tell within a few minutes.

The four habits every person on the team needs

There is a published framework that covers this ground, and it is not tied to any one product. The AI Fluency framework, developed by Rick Dakan of Ringling College of Art and Design and Joseph Feller of University College Cork and released under a Creative Commons license, names four core competencies: Delegation, Description, Discernment and Diligence. Anthropic teaches it in a course built with the same two academics, AI Fluency: Framework & Foundations, which says it teaches people to collaborate with AI systems effectively, efficiently, ethically and safely.

The table below is our working version of those four for a firm that runs on documents. The wording in the right-hand columns is ours.

Competency	The habit, in our words	What it looks like on a Tuesday
Delegation	Knowing what to hand to AI and what to keep	The associate has AI pull the dates, rent steps and options out of a lease, and keeps the judgment about which clause the client needs to hear about
Description	Giving it the context a new analyst would need	The request includes the source document, the firm's template, who the reader is, and what a good one looks like
Discernment	Judging the output against the source	Every figure in the abstract is traced back to a page of the lease before anyone relies on it
Diligence	Owning the result that goes to a client	The person whose name is on the work has read all of it and can explain how it was produced

Two things are worth noticing. Prompting, which is what most training sessions spend their time on, sits inside one of the four (the Anthropic course places its lesson on prompting techniques in the Description module). And the two habits that protect the firm, checking and owning, are the ones a tool cannot supply. We cover the checking step in detail in how to design the human review step in an AI workflow.

Why is the foundation built on real work?

"Before" does not mean a phase with no tools in it. People cannot learn to delegate a task they have never tried to delegate, and they cannot learn to check output they have never seen go wrong.

The foundation is built fastest on one real workflow. Pick something the team does every week, such as a lease abstract, a first pass at a broker opinion of value, or a proposal drafted from past ones. Give the people who do that work a license each. Work through the task together on live documents, with all four habits in play, until the person who owns it is using it without help. Training and the first build overlap, and both finish at about the same time.

What has to wait is everything else: seats for the whole firm, a second and third tool, and builds for teams that have not yet formed the habits. Our guide to running AI training on your team's own work covers the mechanics of those sessions.

What the research says about training and use

The best public evidence on a threshold comes from BCG's 2025 AI at Work survey of more than 10,600 leaders, managers and frontline employees across 11 countries and regions. It found that regular usage is sharply higher for employees who receive at least five hours of training and have access to in-person training and coaching. It also found that only one-third of employees say they have been properly trained.

BCG's 2026 edition, covering close to 12,000 frontline employees, managers and leaders, reports that 74 percent of frontline employees now describe themselves as AI users. Across the survey, only 36 percent of respondents feel they have received adequate upskilling, and 66 percent of frontline employees who use AI regularly get limited or no guidance on what to do with the time they save. Use is widespread, and instruction lags behind it.

These are surveys of what employees report, and they show an association between training and use. They do not prove that five hours causes anything. We read them as support for a floor of about five hours, delivered in person and spread over several weeks, with coaching available in between.

What regulators now expect of firms that deploy AI

One regulator has written literacy into law. Article 4 of the EU AI Act says providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff and of others who operate AI systems on their behalf, taking into account their technical knowledge, experience, education and training and the context of use. The text states that this does not require guaranteeing any specific level of literacy for any individual. That wording comes from the 2026 Digital Omnibus amendment to the Act, which replaced an earlier duty to ensure a sufficient level.

The European Commission's questions and answers on Article 4 confirm that amendment and add three points that matter here. No strict requirements or mandatory trainings are imposed. Simply relying on a system's instructions for use, or asking staff to read them, might be ineffective. And national market surveillance authorities supervise and enforce the article as of August 2, 2026.

This applies to firms within the scope of the Act, which the Commission describes as covering AI systems placed on the Union market, used in the Union, or affecting people located there. A firm with no European clients, staff or operations may well be outside it. We mention it as evidence of direction: a major regulator treats literacy as the deploying firm's job. Whether it or any other rule applies to your firm is a question for your own counsel or compliance lead.

A ten minute readiness check for your team

Ask these five questions in a staff meeting and listen to how specific the answers are.

1. Name one task you handed to AI last week, and one part of that task you kept for yourself. Why did you split it there?
2. What did you give it besides the question, such as a source document, a template, or an example of a good one?
3. Show a recent piece of AI-assisted work. Where in the source document does the second number on the page come from?
4. What client or firm information should not go into the tool, and where is that written down?
5. When AI-assisted work goes out under the firm's name, who has read all of it?

If most of the room answers with specifics, the foundation is in place and more seats or a build will pay off. If the answers are general, or two or three people answer for everyone, the next dollar belongs in the foundation. Question four also tells you whether you need a written policy, and our guide to writing an AI acceptable use policy covers how to make one short enough to follow.

What to buy and build once the foundation is in place

Once the habits hold on one workflow, the sequence is straightforward. Extend licenses to the teams that have been through the same process. Turn the first workflow into something repeatable, with the review step written into it. Then choose the next workflow from what the team learned, since people who have the habits are good judges of which of their own tasks are worth building. Check progress by looking at the work itself, which we describe in how to measure AI adoption on a team without counting logins.

This is how we run our own engagements: we train people inside the build, on their own documents, so the person who owns a workflow is already using it when it is finished. If you are deciding what to approve next, we are glad to talk it through.

Common questions

Should we train the team before buying AI tools?

Buy a small number of licenses so people have something to train on, and build the working habits before you buy seats for the whole firm or add a second tool. The foundation has to come first. A classroom phase with no tools in it does not.

What should AI training for employees cover?

Four habits: deciding which parts of a task to hand to AI and which to keep, giving it the context a new analyst would need, checking the output against the source document, and owning the result that goes to a client. These line up with the four competencies in the AI Fluency framework developed by Rick Dakan and Joseph Feller, which are Delegation, Description, Discernment and Diligence.

How many hours of AI training does an employee need?

BCG's 2025 AI at Work survey found regular usage is sharply higher for employees who receive at least five hours of training and have access to in-person training and coaching. The survey shows an association and does not prove cause. We read it as support for a floor of about five hours, spread over several weeks so people practice on their own work between sessions.

Is AI literacy training legally required?

For firms within the scope of the EU AI Act, Article 4 says providers and deployers of AI systems shall take measures to support the development of AI literacy of their staff. The European Commission says no strict requirements or mandatory trainings are imposed. Whether the Act reaches your firm, and what other rules apply to you, is a question for your own counsel.

Can training happen at the same time as building the first workflow?

Yes, and in our view that is the fastest way to do it. One real workflow gives people a reason to practice all four habits on documents they already know, and the person who owns that workflow is using it by the time it is finished.