The T0–T5 Model: Where Are You With AI?
For: America's Car-Mart Technology Group — corporate IT. Re-anchored for S027 (2026-06-10) from the S025 dealership version.
The big idea
Working with AI is a skill, and like any skill it has levels. We call them T0 through T5.
- T0 = you've never really used it
- T5 = you're orchestrating fleets of AIs that ship work without you
Most of us are somewhere in the middle. This training is built to get every TG employee to T2 by end of session. T2 is where Claude stops being a toy and starts being a real tool — the one that saves hours, not minutes.
You don't need to feel bad about being at T0. The model is descriptive, not judgmental. It's a thermometer, not a grade.
This is the Car-Mart Technology Group training, not the dealership training. Our work spans the API, the data platform, the dealer portal, the runbook library, security, UX, BA, and help desk. The examples below are drawn from those lanes — not from a showroom floor. If you remember an S025 dealership version of this doc, that's the older one. This is the S027 re-anchor for corporate technology.
The tiers
T0 — Confused
One sentence: You've heard about AI tools, maybe tried one once, but you don't really know what they're good for or how to ask.
Example: "I opened ChatGPT once, typed 'help me write an email,' and the response was so generic I closed the tab."
Signal you're here: You avoid AI for work tasks because you don't trust the output and don't know how to direct it.
Why that's fine: Nobody on this floor expects you to show up at T2. T0 is the honest starting point for a lot of TG folks, and the first 30 minutes of the workshop are designed to move you past it.
T1 — Transactional
One sentence: You use AI for one-off questions and rewrites, treating it like a faster search engine.
Example: "I paste a paragraph from a vendor email and ask Claude to make it sound more professional before I forward it to ops."
Signal you're here: You open the chat, ask one thing, copy the answer, close the chat. Each conversation is a one-shot. Every session is a cold start.
What unlocks T2: Real work needs context (your project, your tone, your team's standards). A one-shot chat can't hold that. You need a place to keep the context sticky — and that's a Claude Project in Desktop.
T2 — Project-based (the goal of this training)workshop target
One sentence: You have a Claude Project (or equivalent) for a piece of real work, with files, instructions, and preferences loaded, and you go back to it across multiple sessions.
Example: "I have a Project called api-refactor. It has the 3 most-referenced helper modules, our PR template, and a CLAUDE.md that says 'flag any error-handling that doesn't follow the team pattern.' Every time I open a refactor ticket, I drop the new diff into the Project and ask for a risk summary in our PR template."
Signal you're here: You re-open the same Project multiple times. The AI knows your context without you re-explaining it. The output starts to feel like your work, not generic output.
Why this is the goal: T2 is where the time savings get real. T1 saves you 10 minutes a week. T2 saves you 10 hours a month. Everything above T2 compounds on top of a working Project.
T3 — Multi-domain
One sentence: You run several Projects across different domains, and you know which one to open for which task.
Example: "I have a Project for spec writing, a Project for code review on the dealer-portal repo, a Project for the q3-dashboard data dictionary, and a Project for onboarding new hires. I don't mix them."
Signal you're here: You have 3+ Projects, each scoped to a job-to-be-done, and you instinctively know which Project a question belongs in.
T4 — AI peer / dispatches
One sentence: You treat AI as a peer who can take a well-scoped subtask and run with it; you dispatch to it like you'd dispatch to a junior teammate.
Example: "I told Claude: 'Here are the 47 dealer-service-call transcripts from last week. Categorize them by issue, count each category, and write a one-paragraph summary. Do not invent categories — use only what's in the data.' I came back 4 minutes later to a finished analysis."
Signal you're here: You write prompts that look like delegation messages. You specify the deliverable, the constraints, and the success criteria. You check the output the way you'd check a teammate's work.
T5 — Full orchestration
One sentence: You run multiple AIs in parallel — or have a system that does — where humans stay in the loop on decisions and AI handles the execution and verification.
Example: "I have a setup where one AI drafts the spec, another AI reviews it against our style guide, a third AI generates the test cases, and a human (me) makes the final call. The whole loop runs in under 30 minutes for what used to take a week."
Signal you're here: Your workflow has AI doing real work in the background while you focus on judgment calls. You trust the system enough to walk away from it.
Reality check: Most people — including most engineers at most companies — never need T5. T2 is enough to be dangerous in a good way. T3 and T4 are for people whose job is shipping volume. T5 is for people building the systems other people use.
Where are you? (a 60-second self-check)
Answer honestly. There's no wrong answer, and the answer can change by task.
| Tier | The honest test |
|---|---|
| T0 | I don't have Claude Desktop installed, or I haven't used it for work in the last month. |
| T1 | I use it for one-off rewrites, summaries, or quick questions. I close the tab after each. |
| T2 | I have at least one Project with context loaded, and I've gone back to it more than once. |
| T3 | I have 3+ Projects, each scoped to a different kind of work. |
| T4 | I regularly dispatch well-scoped subtasks and check the output like a teammate's work. |
| T5 | I orchestrate multiple AIs (or have a system that does) with humans in the loop. |
Where you sit right now is fine. The point of the training is to move everyone to T2.
What the path from T1 → T2 actually looks like
Concretely, here's what changes when you cross from T1 to T2.
Open chat → paste the runbook from the wiki → "summarize this" → copy summary → paste into the on-call doc → close chat → next incident, do it all again from scratch.
Open your "Runbook Update" Project (which already has our runbook template, the 3 most-edited runbooks, your team's SRE tone preference, and a CLAUDE.md that says "flag anything about escalation paths or on-call rotation") → drop in the new runbook → ask "compare to the team template and flag deltas" → get a diff-style summary in 30 seconds.
The difference isn't the AI getting smarter. The difference is the AI having your context. That's what a Project is for.
Common questions
"I tried ChatGPT and it was useless. Is this going to be the same?"
Probably not — and here's why. The model isn't magic. The output quality is dominated by the input quality and the context. T1 with no context produces generic output. T2 with the right Project produces work you'd send to a colleague.
"I'm already at T3 or T4. Do I need this training?"
You'll get the most out of the prompts section and the CLAUDE.md section. The T0–T1 content will be a review. Worth your time — please stay engaged so the folks around you have someone to ask when they get stuck.
"I'm a T0. Will I be lost?"
No. The first 30 minutes are built for you. You'll have a working Claude Desktop and a first Project by end of session. We promise.
"Why does this matter for the Technology Group specifically?"
Because the people who get good at this in the next 6 months will be the ones shaping how the rest of the company uses it. You're not just learning a tool — you're becoming the person others ask. Nine directors and CTO Josh are in the room to see what your team can ship today.
"I work in [UX / BA / help desk / security]. Is there room for me in this model?"
Yes. UX personas get to T2 with a Project carrying the latest design system and a CLAUDE.md that says "use our component library, not generic patterns." BA personas get there with a Project carrying the team's spec template. Help desk gets there with a Project carrying the top-20 tickets. Security gets there with a Project carrying the runbook library and a CLAUDE.md that says "flag any suggestion that bypasses authentication." The T0–T5 model is lane-agnostic; the Project content is lane-specific.
What "good" looks like at the end of session
By 4 PM Tuesday, every attendee should be able to do all of these:
- Open Claude Desktop, find their Project, add a file, ask a question, get a useful answer
- Write a prompt that includes role + context + task + constraints (we'll teach the formula)
- State which tier they were at when they walked in, and which tier they're at now
- Name one piece of their real work they're going to apply this to next week
If you can do those four things, you made it. Welcome to T2.
John Whitman — HOOL AI Training, S027 (2026-06-10).
Doc author: Vanguard (S027 v01-t0t5-doc, re-anchored from S025 v01-t0t5-doc). Wire-up: Forge OWM-90.
Spec: S027_AITRAINING_VERCEL_SPEC_v0.1 §5.1. Source of truth: C:/AI/thumbprint/sprints/S027/v01-t0t5-doc/t0-t5-model.md.