Ask any company whether they’ve done AI training for employees and you’ll get a yes. Ask what it was, and watch the answer shrink. It was a webinar. One hour, maybe fifty minutes, called AI Fundamentals or something close to it, and most of the room had it open in one tab while they answered emails in the other.
Somebody posted a photo of the full room on LinkedIn with the word “upskilled” in it. Then Monday came around and everyone did their jobs the exact same way they’d done them the Monday before, except now a couple of people were pasting customer data into a free chatbot on their phone because the paid one needs a ticket to IT.
I get it, I do. The webinar is easy to book and you can put a number on it. Here’s what I can tell you from my side of it though. I’ve got 15 people on my team and all of them work inside Claude all day long, and my own laptop time sits at about 3 hours a week now, down from 90. Not one bit of that came from someone talking at a slide.
It came from sitting next to each person with one real task from their actual week and staying there until what came out was good enough to send. That’s the whole method, and I’ve written it up below so you can run it yourself, brief template and scoring sheet included, along with the three reasons the webinar version is dead within a month.
Why corporate AI training is dead within a month
Corporate AI training falls over for three reasons, and I’ve watched the same three play out at a 40-person agency and a 40,000-person bank. It teaches the tool and skips the job, it happens once, and the only thing anyone measures afterward is who finished the course. Being small doesn’t get you out of this, by the way. It just means the room was smaller.
And the money’s already going out the door. DataCamp asked 500+ enterprise leaders in the US and UK about this in March 2026, with YouGov running the survey, and 82% of them said yes, we offer some kind of AI training. Same survey, 59% of them still say they’ve got an AI skills gap. Only 35% would go as far as calling their program mature and company-wide.
So most companies have the course. More than half of them still don’t have the skill the course was supposed to produce. Sit with that for a second, because it’s the whole problem in two sentences.
Reason 1, it teaches the tool and skips the job
You know the session I mean. What a large language model is, how to write a prompt, where the button is. Now think about your finance manager sitting in it. She didn’t need any of that. What she needed was the month-end variance commentary drafted from the export she already pulls, in her own words, in under ten minutes, and she walked out of that room no closer to it than when she walked in.
That’s the gap behind the number in the TL;DR. Docebo surveyed 2,000 employees and learning leaders across six countries in April 2026, and 85% of the employees said the training they get doesn’t help them use AI in their role. One in five hadn’t been given any AI training at all. Nearly 60% said their company’s learning programs weren’t built with people like them in mind, which is a polite way of saying the course was for someone else.
I mean, would you remember a cooking class that never let you near a stove? Of course you wouldn’t. You’d remember the one where you burned the first pancake and the second one came out right.
Reason 2, it happens once
One 60-minute session, a recording nobody opens, a PDF of 50 prompts. That’s the shape DataCamp’s leaders describe themselves. 40% of them said their most common setup is online learning with the odd instructor-led session, and 24% admitted there are no hands-on projects or labs anywhere in it.
A habit needs two things, repetition and someone watching, and a one-off webinar gives you neither. Gallup’s May 2026 numbers found only 25% of US employees say their company has told them what the plan for AI even is. Where the manager actively backs it, people are 1.7 times as likely to use AI often. So the manager is the training program, whether anyone meant it that way or not.
I know what the slow way costs because I paid for it myself. I spent 18 months building Claude Projects that wrote like a polite stranger before I worked out the Voice DNA layer, which is a 15+ page document that teaches the model how I say things. That was a year and a half of drafts I ended up rewriting by hand anyway, in a business where every draft gets read by a human before it ships. One session with no follow-up gets your people to about week two of that stretch, and then leaves them there.
Reason 3, the only thing measured is who finished the course
Completion rates go up every quarter. The work on people’s desks looks exactly like it did in January. Those two things sit side by side in a lot of companies and no one seems to find it odd. DataCamp found 26% of leaders can’t measure the return on their training at all, so the dashboard shows a green tick next to every name and the variance commentary still takes an hour.
Ask the employees instead of the dashboard and the trend runs the wrong way. Jobs for the Future surveyed 3,020 workers and published in March 2026 that 36% feel they have the training and resources they need to use AI, down from 45% the year before. Then PwC published findings in September 2026 from 49,364 workers across 48 countries and found 64% had used AI at work in the past year, up from 54%, while the share who can get the learning they need slipped from 59% to 51%.
So more people are using it and fewer people are being taught how, in the same twelve months. That’s the paradox every corporate AI training budget is paying for right now.
The six-week shape of a program that sticks
Here’s the shape, and I’d keep it exactly like this before you change a single thing. Six weeks. One live session a week, 90 minutes long, with two hours of practice in between on the person’s own work, the stuff already sitting in their inbox that week. And their manager is in every session doing the same homework as everyone else, which I’ll come back to, because it’s the piece that gets dropped first and matters most.
One thing before week one though. Run the 24-question readiness check first, so you know which teams have a process worth training on and which ones still need theirs written down. I say that because you can’t hand a task to AI while the task only exists in someone’s head, and more teams are in that spot than will admit it on the call.
| Week | Live session, 90 minutes | Homework, 2 hours | The manager’s job |
|---|---|---|---|
| 1 | One tool, one login. Everyone opens a paid seat and runs their first real task live. | Run the same task three more times on three real inputs. | Pick the one recurring task each person will train on. |
| 2 | Writing the brief. Each person drafts the one-page brief for their task with three examples of their best past work. | Finish the brief. Run the task from it. Note what came out wrong. | Read every brief. Send two back. |
| 3 | Reviewing like an editor. The 12 tells, the humanizing pass, the ship-or-fix call. | Review five AI drafts against the brief. Fix each one in place. | Run the first cold task. Keep the scores. |
| 4 | Building the Project. Brief, examples and rules saved in one place the person opens every morning. | Run the whole week’s version of the task from the Project only. | Check the Project exists and gets opened. |
| 5 | Edge cases. What the Project gets wrong, what stays human, where the handoff line sits. | Write the three things the Project must never do for this task. | Sign off the handoff line in writing. |
| 6 | The cold task, the review test, and a show and tell of one finished piece per person. | None. The habit is the homework from here. | Put the 30-day check on the calendar now. |
The curriculum, four modules per role
Teach four things per role, in this order: one tool and one login, the brief, the review, the Project. Everything else waits until after week six, including agents, automations and anything with the word “workflow” in the title.
Module 1, one tool and one login
Pick one assistant for the whole company and buy everyone a paid seat. We run on Claude, and I wrote up the 30 marketing tasks I ran through both if you want the reasoning. Whichever you pick, the rule is one tool, so the briefs and Projects your people build in week two still work on every desk in week twenty.
Personal accounts come off the table on day one. A paid seat people are allowed to use closes the free-chatbot-on-the-phone problem from the intro faster than any policy document, because the policy was never the obstacle. The ticket to IT was.
Module 2, the brief
The brief is one page about one recurring task, written by the person who does the job, in their words, with their manager reading it back to them. My own version is the Voice DNA document sitting on top of 100,000+ words of my own writing. Your finance manager’s can be one page, built from the five boxes below.
| Box | What goes in it | The finance manager’s version |
|---|---|---|
| 1. The task | One sentence, one recurring job | Draft the month-end variance commentary for the leadership deck |
| 2. What goes in | The exact inputs, where they come from | The ledger export (CSV), last month’s commentary, the budget sheet |
| 3. Who reads it | The real reader and what they do with it | The CFO, in a 20-minute meeting, to decide which three lines to ask about |
| 4. What good looks like | Three examples of this person’s best past work, pasted in full | The March, June and September commentaries she was proudest of |
| 5. Never | The lines the output must not cross | Never guess a cause for a variance. Never round to the nearest thousand. Never open with “Overall” |
This is the module people skip because it feels like paperwork. It’s the one that decides whether the output sounds like your company or like a press release from 2019.
Box 5 does more work than the other four put together, so make people write it last and argue about it longest.
Module 3, the review
Every draft gets read by a human before it ships. I’ve kept that rule through twelve months of running AI staff on my own business. The review module teaches people to read like an editor, which means catching the 12 tells that give AI writing away and running the 7-step humanizing pass on anything a customer will see.
Fix the draft in place. The moment someone rewrites an AI draft from scratch they’ve stopped training the model and started doing the job twice. The review habit is where the time saving lives, so it gets a whole week.
Module 4, the Project
A Project is a saved space inside the assistant that holds the brief, the examples and the rules, so the task runs the same way every morning without anyone re-explaining it.
Claude calls them Projects. Other tools have an equivalent. Your people build one each in week four, paste the five boxes from module two into its instructions, and open it every day after that.
Once a Project has run the same task cleanly for a month, it’s a candidate for AI business automation, where the human moves from doing the task to reviewing its output. That’s a different post and a different budget. Six weeks in, the goal is one Project per person that gets opened on Monday. If the task is website copy, the 12 LLM SEO signals are the next check to run on what the Project turns out.
| Role | The one recurring task for the brief | What ships at week six |
|---|---|---|
| Sales | Follow-up email after a discovery call, from the call notes | A Project that drafts the follow-up in the rep’s voice in under five minutes, reviewed and sent the same day |
| Finance | Month-end variance commentary from the ledger export | First draft ready the morning after close, reviewed by the manager |
| Marketing | Turning one long piece into four channel-specific posts | Four drafts per piece with the 12 tells checked before scheduling |
| Support | First reply to a ticket in the three most common categories | Draft replies in the brand tone, approved before sending |
| Operations | Weekly status update from the project tracker | A status draft the ops lead edits in ten minutes |
The cadence, 90 minutes a week with the manager in the room
Ninety minutes live, once a week, for six weeks, is the minimum worth scheduling. Anything shorter gets nodded through and forgotten by Thursday.
The manager attends every session and does every piece of homework on their own task. This is the part of AI upskilling that gets negotiated away first, usually with the line “I’ll catch the recording.”
Gallup’s numbers say the opposite. Where the manager actively supports the tool, frequent use is 1.7 times as likely, so the manager’s seat in the room is worth more than any slide you’ll show.
Expect resistance in week two and quiet in week four. Week two is when the brief feels like homework. Week four is when the Project starts working and people stop asking questions because they’re busy using it.
If week four is loud, something in the brief is wrong, so go back and read it with them.
The wider numbers tell you how far behind the average company sits. BCG’s June 2026 AI at Work survey of 11,749 people found 36% feel they received sufficient training, the same share as in 2025, while 88% expect to need major upskilling in the next five years.
Your people already know they’re under-trained. Six weeks of showing up is the whole gap.
The assessment, three tests that tell you it worked
Employee AI enablement gets measured on output. Three tests, run at fixed points, tell you whether the six weeks landed: the cold task, the review test and the 30-day check. None of them involves a certificate.
The cold task
At week three and again at week six, each person gets a fresh version of their recurring task, 20 minutes, no help. Their manager scores the output on the sheet below against the same bar they’d hold a human draft to.
Week three scores will be rough. Week six scores should be shippable with light edits. The gap between the two is your training result in one number.
| Criterion | 0 | 1 | 2 | 3 |
|---|---|---|---|---|
| Matches the brief | Ignored it | Right task, wrong inputs | Right inputs, one box missed | All five boxes honored |
| Sounds like the person | Generic | Right facts, wrong voice | Close, two tells left | The manager couldn’t tell |
| Ready for the reader | Needs a rewrite | Needs heavy edits | Light edits | Could send as is |
| Time to done | Over 20 minutes | Under 20 | Under 15 | Under 10 |
Twelve points available. Treat anything between 4 and 7 at week three as normal. Anyone at 10 or above by week six is ready to run their Project unsupervised. Anyone under 8 repeats module two, because a low score almost always traces back to a thin brief and almost never to a slow learner.
The review test
Hand each person an AI draft with five problems planted in it. One invented figure, one generic opener, one sentence that contradicts the brief, one tone slip, one claim with no source.
Four out of five caught is a pass. Anyone catching two or fewer repeats module three before they’re allowed to ship anything unsupervised.
The 30-day check
Thirty days after the last session, pull the usage report from the tool’s admin console and count how many people opened their Project in each of the four weeks. Set the pass mark at eight in ten before you look.
For context, Gallup’s May 2026 wave puts daily AI use across all US employees at 15% and use a few times a week or more at 30%, so eight in ten weekly users inside one team is a long way above the national picture. It’s also the only number that proves the habit survived the launch memo.
Who runs it and what it costs
Someone who already does this every day runs it. That rules out most L&D teams through no fault of their own, since the person leading week three needs to have reviewed a few thousand AI drafts themselves to know what a planted problem looks like. If you have that person in-house, give them six weeks and this post.
If you don’t, this is the work my team does in Done-with-You, $15,000 over 2 months, where we build the first Projects alongside your people and run the sessions with your managers in the room. Done-for-You at $40,000 over 2 months is for companies that want the Projects built and handed over with the training wrapped around the handover. I’ve written separately about what AI consulting costs across the market and how to measure the return. Hold us to both.
| Route | What you get | Cost |
|---|---|---|
| In-house | One experienced person running six weeks, paid seats on one assistant | Their time, plus the seats |
| AI Labs | 60+ courses, two live implementation workshops a week, the four modules to run yourself | $197 a month per person |
| Done-with-You | My team builds the first Projects alongside your people and runs the six sessions | $15,000 over 2 months |
| Done-for-You | Projects built and handed over, training wrapped around the handover | $40,000 over 2 months |
For a solo founder or a team of three, the full program is overkill. AI Labs is where 5,000+ students have been trained so far. Running yourself through the four modules there is the cheapest way to find out whether the whole team needs the Done-with-You version.
Your team is already asking for this
Let me ask you this. When did an employee last ask you for more compliance training?
Now count how many have asked, this year, for help using AI properly. The appetite is already in the building, it’s the Chief AI Officer’s auditorium that keeps wasting it.
Your people are waiting for a program that treats their job as the subject, which is a far easier ask than convincing them AI matters.
Honestly, it’s mainly a scheduling decision at this point. Six Tuesdays at 90 minutes, one real task per person, and a team working the way mine does at the end of it.
FAQ on AI training for employees
What should AI training for employees include?
AI training for employees should include four modules built around each person’s own recurring task. One tool with a paid seat, a one-page brief written by the employee in their own words, a review habit that catches AI tells before anything ships, and a saved Project that runs the task the same way every day. Generic sessions on what a language model is can be cut entirely.
How long does AI training for employees take to stick?
AI training for employees takes about six weeks to become a habit when it runs as one 90-minute live session a week plus two hours of practice on real work. A single session, however good, fades within a month. The test of whether it stuck is a usage check 30 days after the last session, with eight in ten people still opening their Project weekly as the bar.
Why does corporate AI training fail?
Corporate AI training fails for three reasons. It teaches the tool and skips the person’s job, it happens once with no manager follow-up, and it measures course completions when it should be measuring who can do the task afterwards. In Docebo’s April 2026 survey of 2,000 employees and learning leaders, 85% of employees said the training they received does not help them use AI in their role.
How do you measure employee AI enablement?
Employee AI enablement is measured on output from three tests. A cold task at week three and week six, scored out of 12 by the manager on brief match, voice, readiness and time. A review test where the employee has to find at least four of five problems planted in an AI draft. A 30-day usage check that counts who still opens their saved Project every week once the sessions end.
Should employees learn Claude or ChatGPT?
One of them, chosen for the whole company. Employees learn faster when every desk runs the same assistant, since the briefs and Projects transfer between people and teams. First Movers AI runs on Claude for writing-heavy work, which is where most office tasks sit. As of November 2026, all 15 people on the First Movers AI team work inside it all day. Either tool works for a six-week program as long as it is paid, shared and the only one in use.
How much does AI upskilling for a team cost?
AI upskilling for a team costs $197 a month per person for a self-serve membership such as AI Labs, $15,000 over 2 months for a Done-with-You engagement with First Movers AI where the Projects get built alongside your staff, and $40,000 over 2 months for Done-for-You. The in-house version costs one experienced person’s time for six weeks plus paid seats on one assistant.
Run the six weeks with us
Say it’s week seven. Your finance manager opens her Project on the first morning after close, drops in the export, and reads a draft of the commentary that sounds like her. She fixes two lines and sends it before her coffee’s cold. That’s what the six weeks buy, repeated across every desk that has a recurring task on it.
If you want my team running those six weeks inside your company, with your managers in the room and the first Projects built alongside your people, book a call with my team. We take a small number of companies each month. If you’d rather start on your own, AI Labs has the four modules waiting and two live workshops a week to ask questions in.
Six Tuesdays will do more for your team than every webinar they’ve ever sat through on mute. None of it needs a LinkedIn post.
