How do you run corporate AI training that actually sticks?
Short answer
Lead with mindset, not tools. Make it hands-on with laptops open and real work, standardize the company on one AI platform, and teach individual use before any shared team system. I run it as four touchpoints: an intro call, a prep call the week before, the in-person workshop, and a follow-up one to two weeks later.
A corporate client told me a nine-person cohort from their AI training saved about 2,700 person-hours a year within two months. That's their number, not mine. The format behind it is the one I'd copy.
What should AI training lead with?
Mindset and curiosity, not tool tutorials. An HR director came to me whose company paid for both ChatGPT and Claude, and team adoption was near zero. The tools weren't the problem.
"Mindset shift comes before tools. Get people to understand what AI chat can already do for them as individuals before introducing any shared multiplayer team system, which is still unsolved industry-wide."
Being willing to change when things change around you is one of the hardest things for most people. That's exactly why so many don't want to use AI. They're not used to it.
What format works?
The client with the 2,700-hour result scoped a repeat of a format that had already worked for them:
- A full day in person
- One lead trainer plus one support person
- Change-management framing first
- No hierarchy in the room
- About 90 minutes at the end with each attendee working a real business use case
Starting with a small cohort created organic demand inside the company instead of fear. A weekly show-and-tell afterward keeps people sharing wins.
How should the engagement be structured?
For the HR director I proposed four touchpoints:
- An intro Zoom
- A prep Zoom the week before
- A hands-on half day in person, laptops open, building real things
- A follow-up Zoom one to two weeks later
Pull the use cases from their actual week: interview scheduling automation, cross-referencing documents, repetitive admin.
Which AI platform should the company standardize on?
Pick one and stay on it. In my assessment in mid-2026, ChatGPT was ahead of Claude for general use, even though business media coverage favored Claude. Either way, a company shouldn't keep switching platforms. Teach general use first, then layer in shared infrastructure.
Expect IT to be cautious. That HR director's IT team was worried about data corruption. Starting with individual use helps there too.
What should I charge for a first full-day training?
More than you think. My first full-day inquiry came from a 300-person company that only wanted to train 10 people, to really dig in for a whole day. When she asked how my other trainings had gone, I told her the truth: I'd never done a full-day training. I threw out $5,000 anyway, and she didn't bat an eye. She wrote back with exactly what they were looking for.
The scary part wasn't the price. The people she was sending weren't marketers. They were an accountant, the head of acquisitions and the head of accounting. Those are use cases I'd never worked, so I had to learn them before the day. She also expected a second person in the room to help attendees.
Should I teach the technical tools?
Not for most teams. I use Codex, and you don't need the terminal. You just do it in Codex and it does everything. I don't think it makes sense to teach people something that's going away, when they can work in Codex, Cowork or a desktop app. Unless someone is a true developer, skip the terminal.
Is corporate AI training a good business?
Yes, though not because of the training fee. Corporate AI training is a network play. The first engagement leads to implementation, consulting and referrals.
Know your limits, too. When a 300-person client wanted a structured two-presenter day that included Microsoft 365 integration, I went looking for a co-trainer with corporate experience.
For the public version aimed at owners instead of employees, see how to run an AI workshop for business owners.
Answered by TJ Larkin. From client and partner calls in May, June and September 2026. The cohort result is a client's own report. Last updated .