The short answer
Build an AI training program in layers: shared foundations (clear requests, review, data rules), role tracks built on each team’s real workflows, tool-specific modules only where a tool is approved, and a reinforcement system of champions and office hours. Govern with a short policy and measure behavior change against a baseline.
What does a complete AI training program include?
A complete program has six parts: a shared foundation, role-based tracks, tool-specific modules, governance, reinforcement, and measurement. Programs that fail usually have only the first, taught as a single generic session.
The big shift in 2026 is that most organizations have more than one AI tool. Chat assistants, work agents, and coding agents from more than one vendor all exist at once. The skills that transfer, such as asking clearly, bounding tasks, and reviewing output, should be taught once, with tool-specific modules layered on top.
How should the levels be structured?
Use three levels with clear entry points.
| Level | Audience | Focus |
|---|---|---|
| Foundations | Everyone | Safe daily use: requests, context, review, data rules |
| Workflow builder | Regular users | Saved instructions, shared Projects or equivalents, connectors, templates |
| Agents and automation | Power users, champions, developers | Delegated multi-step tasks, scheduled work, skills, coding agents |
This mirrors how we structure the programs we deliver: Claude Team Kickstart, the Enterprise Claude Rollout, and the Enablement Partnership. The approach page shows how the levels fit a rollout.
Which tool-specific modules do you need?
Add a module only for tools you have approved. Typical modules:
- Claude: Projects, Skills, connectors. See Claude training for teams.
- ChatGPT: modes, saved workflows, admin controls. See ChatGPT training for employees.
- Work agents: Claude Cowork and ChatGPT Work: delegation, permissions, scheduled tasks.
- Coding agents: Claude Code and Codex: instruction files, permissions, review.
Do not train people on tools they are not allowed to use, because it invites shadow use.
How do you govern it?
Write a short acceptable-use policy and train on it with scenarios. Cover approved tools, data rules, review, agents and connectors, and incident reporting. Our policy guide gives a section-by-section outline, and the governance kit provides supporting materials.
Have legal and security approve the final version. This is educational guidance, not legal advice.
How do you make it last?
Use champions and office hours. A trained champion in each department answers questions faster than a ticket queue and tells you what is confusing. Hold office hours two weeks after each session, where people bring what broke. Maintain a shared library of the instructions and workflows that worked.
Name a program owner. Training without an owner decays when the calendar invitation ends.
How do you measure results?
Measure three layers. Learned: a scenario check that tests judgment. Using: active use over time by role, from your admin analytics. Improved: one or two task-level measures with a baseline recorded in advance.
Report what you measured, how, and over what period. For a worked example of a large, regulated rollout, see our enterprise rollout case study. To find a sensible first use case for your organization, take the Claude Readiness Check.
How do you handle shadow AI in an employee training program?
Shadow AI is employees using AI tools you have not approved, usually because the approved path is unclear or slower. Banning it rarely works. A better approach is to make the sanctioned path easy:
- Publish the approved list and how to request a new tool.
- State the data rules in plain language with examples.
- Give people something better: role workflows that save real time inside approved tools.
- Make reporting safe so mistakes surface early.
- Measure and adjust: survey use, review requests, and update the list.
Training is where these habits form, so teach the rules on real tasks, not slides. Our acceptable use policy guide covers the policy side.
How do role-based and regulated-industry programs differ from generic AI courses?
A generic course explains what AI is. A role-based program changes the weekly work of a specific function, with that function’s tasks, tools, and review owners. In regulated industries the program also has to teach data classification, disclosure, record-keeping, and when a human must approve.
If you want to compare formats, ask each option: whose tasks do we practice, whose rules do we teach, who supports people afterward, and how will we measure use? Our programs are built around those four answers, and our approach page explains how.
Frequently asked questions
How long does an AI training program take?
Plan for a phased program over weeks, not a single day: foundations first, then role tracks, then advanced content and office hours.
Should we train on one tool or several?
Teach shared skills once, then add modules for each approved tool.
What does it cost?
It depends on scope. Our programs are listed on the programs page with starting prices.
How do we get leadership buy-in?
Start with a bounded pilot, a baseline, and a short written policy. Show results with honest numbers.
Can you train a regulated team?
Yes, with data rules defined first and compliance involved from the start.
What is shadow AI?
It is employees using AI tools the organization has not approved. The best fix is a clear approved list, plain data rules, and workflows that make the approved path faster.
How do you measure AI training success?
Track active use of approved tools, time saved on defined tasks, review error rates, and policy incidents against a baseline you set before training.
Sources and further reading
Product features, plans, and limits change often. This guide reflects the vendor documentation listed above as of November 2026. Check the vendor’s current documentation before you make purchasing or policy decisions. TechStack Consulting is an independent training firm and is not affiliated with Anthropic or OpenAI.