Choosing an AI training provider for a team comes down to one question with twelve parts: what will the team be able to do, and still be doing, a year after the last session? Providers fall into four formats (recorded libraries, single workshops, public cohort courses and custom programmes), and each answers that question differently. This guide gives you the twelve questions TAI Labs would want to be asked, what a good answer sounds like, and how to compare formats honestly, including when we are not the right choice.
The four formats, compared honestly
| Format | Suits | What the team keeps |
|---|---|---|
| Recorded library with a seat licence (Coursera for Business, LinkedIn Learning, DataCamp and similar) | Large headcounts needing a baseline of literacy at low cost per seat | Completion records and general knowledge; nothing specific to your workflows |
| Single workshop, half a day to a day (many agencies and consultancies) | A team that wants an introduction and a good afternoon | A deck, a prompt gallery, some enthusiasm |
| Public cohort course (Maven, Section, Reforge and similar) | Individuals who want a strong course on a topic | Trained individuals with no shared team system |
| Custom programme on your own workflows (TAI Labs, and some enterprise providers) | A team of 5 to 50 with a repeated workflow to change | Working workflows, a skills library, a knowledge and training system, a readiness report |
The twelve questions to ask any provider
Will you train on our workflows, or on yours?
A good answer names the discovery step: interviews with the people doing the work, before any syllabus. A bad answer is a course catalogue.
Who teaches, and what do they ship?
Ask for the instructors by name and what they build today. Practitioners who ship with the tools can answer the question your team will ask in the lab. Curriculum-only trainers cannot.
Which tools, specifically?
Expect named tools and a reason for each: Claude and ChatGPT for briefs, Claude Code or Codex for engineering, n8n or Make for operations, custom MCP connections for your data. 'AI tools' is not an answer.
Is the work done in the sessions, or watched?
Live labs on your own material, with an instructor beside each person, or a recording. A quarter of enterprise leaders say their AI training lacks hands-on labs; do not add to the number.
How is capability assessed?
Assessed work (a workflow that runs, a skill file, a brief a reviewer accepted) scored against a rubric. Attendance certificates measure attendance.
What baseline do you take, and what do you compare it to?
Real tasks timed and graded before the first session, repeated like for like at the end. If the provider cannot describe this, it cannot report a return. The ROI guide has the method.
What does the team keep?
Skill files, prompt libraries, SOPs, connections with their permissions written down, session recordings, and the programme itself on your learning platform. A deck is not a system.
What happens with our data?
Which tools receive it, under which agreements, whether anonymised or sample data can be used, and who reviews outputs before they leave the team. Security should be able to sign this off from the answer.
Do you need a retainer afterwards?
The honest answer is office hours for a quarter and then nothing, unless you want a new capability taught. A monthly coaching retainer means nothing was handed over.
Can our own people run the next intake?
A programme that ends with champions and a handed-over curriculum can be run again by your team. One that depends on the vendor's slides cannot.
Show us one team like ours
A case study with the team size, the format, the workflows and what was produced, not a percentage with no method behind it. Read TAI Labs' customer stories with that test in mind.
What does it cost, and what moves the number?
Expect a format, an indicative range and the factors that change it, then a written proposal after scoping. The cost guide shows what that looks like from us.
The tools a serious provider teaches on
The tools change every quarter, so the list matters less than whether the provider names them and can teach the connections between them. These are the ones TAI Labs teaches most in 2026.
ClaudeBriefs, analysis and reusable skills
ChatGPTResearch and everyday work
Microsoft CopilotTeams that live in Microsoft 365- Claude CodeEngineering, and encoding processes as skills
CodexCoding agents and background tasks
CursorAgent-assisted work in the codebase
n8nConnecting operations workflows
MakeVisual automation for approvals
HiggsfieldImage and video concepts for marketing
Custom MCPsApproved connections to your own systems
When TAI Labs is not the right choice
We are a small practitioner team, and the format we run is built for teams of roughly five to fifty people with a repeated workflow to change. If you need a literacy baseline for two thousand people this quarter, a recorded library with a seat licence is the right buy and we will say so on the call. If one person wants a strong course, a public cohort is better value than a custom programme. If your security policy forbids any company data in any AI tool, start with the policy, not the training.
- Thousands of seats needing basic literacy fast: a recorded library
- One or two individuals: a public cohort course
- No approved AI tools yet: policy and tool selection first, then training
- A team of 5 to 50 with a workflow to change and a leader who wants it: a custom programme
Do it with us once
Ask us the twelve questions.
A 15-minute call, a real person, no deck. We answer every question above for your team and send the outline whether or not you buy.
A short checklist for the shortlist
- Discovery on our workflows before any syllabus
- Named practitioner instructors
- Named tools, with the connections between them taught
- Live labs on our own material
- Assessed work, a baseline and a like-for-like comparison
- A handover: skills, SOPs, recordings, the programme on our platform
- No retainer required; our people can run the next intake
- A case study with a method behind it
Frequently asked questions
Should we buy a platform or a programme?
A programme changes how a team works; a platform keeps the change alive and measured across the company. Most companies run a programme first, then decide about a platform with evidence. TAI Labs offers both, and the programme plus platform option combines them.
How do we compare prices across formats?
On what the team keeps, not on price per session. A single workshop that gets bought twice costs more than a programme bought once. The cost guide sets out the ranges by format.
Are certifications worth anything?
An assessed certification, scored on work rather than attendance, is evidence a manager can check. A completion certificate is not. Ask how the assessment is done.
How do we run a pilot?
A workshop series on one workflow with one team: discovery, two live sessions, labs between them. Set the baseline first. The 90-day rollout guide shows how a pilot becomes a company-wide rollout.
What makes TAI Labs different?
Discovery on your workflows, live labs with an ML PhD and an exited AI founder, assessed work with a readiness report, and a handover that includes the knowledge and training system so you do it once. The custom AI training page has the formats, instructors and customer stories.
Decide what the team must be able to do a year from now, ask every provider the twelve questions, compare formats on what the team keeps, and choose the recorded library, the public course or the custom programme accordingly. If it is the custom programme, that is the one TAI Labs runs, once.