Give ten people access to the same AI tools and come back in 90 days. You will not find ten versions of the same thing. You will find three fundamentally different relationships with the technology; and only one of them compounds the way you hoped.
Every team given AI access develops these three distinct patterns within 90 days. The founder who can identify which pattern each person is running can stop managing AI adoption by vibe and start managing it by data.
The three types — defined in operational terms
Each type is described around what they actually do on a Tuesday; not who they are as a person. This is a recognition tool, not a performance review.
Type 1 — The Cyborg
AI is woven into every part of how they work. They do not open a task and then decide whether to use AI. They open AI and then decide what to work on. The context for their work is already loaded somewhere; a shared project, a personal workspace, a saved prompt library. They move faster than anyone else on the team and produce outputs that sound like the company because they have built the context layer themselves.
What Tuesday looks like: Opens shared AI workspace first. Draft email: AI. Meeting prep: AI. Summarise last week’s notes before a client call: AI. They are not “using AI tools”; they are working inside an AI environment.
The risk: Their system lives in their head and their personal workspace. When they leave or take a holiday, the system degrades because nobody else knows how they run it.
Type 2 — The Centaur
Half human, half AI; but the seam is visible. They use AI for specific tasks they have decided AI is good at: drafting emails, summarising documents, generating first-cut reports. For everything else, they work the old way. The boundary between “AI tasks” and “my tasks” is explicit in their mind and they defend it.
What Tuesday looks like: Opens their inbox. Drafts a standard client email with AI. Writes the internal strategy memo by hand because “AI doesn’t understand the context.” Runs the financial summary through AI. Prepares the board presentation manually because “it needs to sound like me.”
The risk: They are leaving significant leverage on the table. Not because they are lazy or resistant; because they have not seen a workflow that convincingly handles the tasks they have reserved for themselves. The gap is demonstration, not persuasion.
Type 3 — The Self-Automator
They use AI to build things. Not just to assist with tasks; to replace them entirely for themselves. They have automated their own reporting, their own email triage, their own research process. They are living six months ahead of the rest of the team in terms of what AI can do. And they have documented none of it.
What Tuesday looks like: Their weekly report writes itself from a prompt they built three months ago. Their inbox is triaged before they open it. They finish by 3pm what used to take until 6pm. When a colleague asks how, they say “oh it’s just a thing I set up” and cannot easily explain it.
The risk: They are the most valuable AI asset in the company and the most fragile one. Their automation lives in their personal account. It has no documentation. It has no successor. It compounds only for them.
How to identify which type each person on your team is
Three observable signals for each type. The founder should be able to run this assessment by watching how the team works for one week; not by asking them.
Signals of a Cyborg:
- Their output quality is consistently higher than the baseline; emails more polished, reports better structured, turnaround faster
- When you ask “how did you put this together?” the answer usually involves AI at multiple points, not just one
- They proactively share prompts or workflows with colleagues without being asked
Signals of a Centaur:
- Their output is inconsistent: excellent in some areas (usually the ones they have automated), standard in others
- When you review their work, the AI-assisted sections are visibly different in quality and style from the non-AI sections
- They use AI reactively (“let me run that through AI”) rather than proactively (“here’s what I built”)
Signals of a Self-Automator:
- They finish certain categories of work unusually fast compared to peers in the same role
- They are vague when asked how (“I have a system” or “I set something up”)
- Their personal productivity is high but their team’s productivity has not improved since they joined
The one-question team audit:
Ask each person: “Walk me through the last piece of work you completed this week that involved AI. What did you put in, what came out, and is that process documented anywhere?”
| Answer type | Likely type | What it reveals |
|---|---|---|
| Detailed, multi-step, references shared tools | Cyborg | High adoption; check whether it is documented |
| One specific task, AI was a tool not a system | Centaur | Selective adoption; gap is in workflow coverage |
| Vague or deflects (“it’s just a quick thing”) | Self-Automator | High capability; zero documentation; fragile |
| Has not used AI this week for anything material | Non-adopter | Different intervention needed; outside this framework |
What each type needs from you — the specific intervention for each
This is a management playbook, not an HR strategy. One concrete action per type.
For the Cyborg — extract and systematise
The Cyborg’s value is not just their personal productivity. It is the working AI practice they have built. Your job is to get that practice out of their personal workspace and into the shared system.
The intervention: schedule a 90-minute session with them. Ask them to walk through a typical Tuesday and show you exactly what they run and how. Document every workflow they demonstrate. Load the context they use into the shared workspace. The goal is not to slow them down; it is to make their practice the company’s practice.
The risk if you skip this: They leave and take the company’s best AI system with them.
For the Centaur — demonstrate, do not mandate
The Centaur is not resistant to AI. They are resistant to AI that has not proven itself in the tasks they care about. Telling them to use AI more does not work. Showing them a workflow that handles one of their reserved tasks convincingly does.
The intervention: identify one high-friction task they currently do manually (the strategy memo, the board presentation, the client brief). Build a working AI workflow for that specific task using the company context pack; not a generic demo, their actual task with their actual context. Run it with them once. Let the output quality do the persuading.
The risk if you skip this: The Centaur plateau is permanent. They stay useful but never compound.
For the Self-Automator — document or lose it
The Self-Automator is building the company’s most valuable AI infrastructure without being asked and without documenting any of it. This is not a performance problem. It is a knowledge management problem.
The intervention: give them a formal role in the AI system. Make them the workflow owner for their department. Ask them to document every automation they have built; inputs, outputs, prompt structure, where it lives; into the shared workspace. Frame it as a promotion, not a request. The Self-Automator builds things; giving them an official system to build into channels their instinct productively.
The risk if you skip this: Three months from now they move on, and so does everything they built.
Why people move between types — and how to accelerate it
The most common reason a Centaur does not become a Cyborg is not personality. It is that the AI environment they work in makes going deeper harder than staying where they are.
If there is no shared context pack, every new task requires rebuilding context from scratch. If there are no documented workflows, they have to invent the wheel every time. If there is no adoption tracking, nobody notices when they use AI more or less.
Three infrastructure changes that move people between types:
- A shared context pack reduces the friction of going deeper. When the company voice, customer archetypes, and decision rules are pre-loaded, the Centaur does not have to front-load context for every new AI task. The barrier to trying AI on a new task drops significantly.
- Documented workflows give the Centaur an on-ramp into tasks they previously reserved for themselves. They do not have to invent a workflow for the strategy memo. They run the documented one, see it works, and expand from there.
- Adoption tracking creates accountability without surveillance. When the team knows that workflow usage is visible; and that high adoption is recognised; the social dynamic around AI use shifts. Using AI more becomes the visible norm rather than the invisible exception.
The team distribution most mid-market companies actually have
Based on Phos engagements at the point of starting work with companies at $5M–$25M:
| Type | Typical share of team | What this means |
|---|---|---|
| Cyborg | 5–15% | Usually 1–3 people in a 20-person company; often includes the founder |
| Centaur | 40–60% | The majority; using AI but not compounding |
| Self-Automator | 10–20% | Present in almost every team; usually invisible |
| Non-adopter | 20–35% | Not the problem to solve first |
A well-functioning AI team at 12 months post-Foundation build typically looks like: 25–35% Cyborgs, 50–60% Centaurs running documented workflows, 10–15% Self-Automators whose work is now documented and shared.
The mistake most founders make: trying to move Non-adopters first. The Non-adopter problem largely solves itself when the Centaurs around them start visibly producing better work faster. Social proof inside the team is more powerful than any training programme.
Common questions on managing AI adoption by type
”What if my whole team is Centaurs?”
That is actually the most common finding at companies that have given everyone AI access but have not built shared infrastructure. Centaurs are not a failure state; they are a starting point. The move from Centaur to Cyborg is faster when the shared context pack exists and documented workflows are available. Fix the infrastructure first; the distribution shifts on its own.
”Can a Non-adopter become a Cyborg?”
Sometimes; when the social norm shifts enough and the infrastructure makes it easy. Mandate-based approaches rarely work. The Centaurs and Cyborgs in the team demonstrating better outputs is what usually moves Non-adopters; not a policy.
”Is the Cyborg pattern sustainable long-term?”
Yes; if their practice is documented and shared. The unsustainable version is a Cyborg who has built a private system with no documentation. When that person leaves, the system degrades. The sustainable version is a Cyborg whose practice has been extracted into the shared workspace; so it compounds for the company, not just for them.
”What if the Self-Automator resists documentation?”
Frame it as ownership, not compliance. “We want to make sure this keeps running when you’re not here, and we want to build more infrastructure like what you’ve built” lands differently than “we need you to document your processes.” The Self-Automator is motivated by building. Give them a system to build into.
”Does this framework apply to senior leadership?”
Yes; and the stakes are higher. A COO who is a Self-Automator has built operational automation that the business depends on and that nobody else can maintain. A senior leader who is a Non-adopter has more influence over whether the rest of the team adopts or not. Map the leadership team first.
”How do I track adoption type without it feeling like surveillance?”
The adoption tracking signal is about workflow usage, not individual monitoring. “The shared pipeline summary workflow ran 12 times this week” is system data, not surveillance. When the framing is about system health rather than individual performance, the team generally accepts it. Frame it correctly from the start.
Want to know exactly which type your key people are — and what to do about it?
If the 30-minute audit produced a long “No” column, that is the common finding. Most founders at $5M–$25M have an impressive personal AI practice sitting on top of an invisible infrastructure problem.
Path one: start this week. Write the voice guide first. Two hours, a Google Doc, no approval required. That one document, loaded into a shared workspace, will immediately lift the floor on every AI output your team produces.
Path two: bring in a partner. If mapping the team, extracting the Cyborg’s practice, and installing the shared infrastructure needs to happen in weeks rather than months; that is the work Phos does. The fastest way to know if it’s the right fit is a conversation. Thirty minutes, no deck. Start here.