A diagnostic and coaching guide for the five stages of AI adoption, because most companies are two levels behind where they think they are.
Most companies believe they are further along than they are. But perception doesn’t always match reality.
And that’s where the competitive advantage is won or lost.
Ask a founder how their team is doing with AI and you’ll hear some version of the same answer. “I use ChatGPT to draft emails.” “We have a share prompt in Notion.”
Okay. Now remove every AI tool from your team. If the business runs the same way, just a bit slower, nothing has actually changed. Most teams are in that category and don’t know it.
There are five levels of real AI adoption. Most companies are sitting at Level 1 or 2 while telling themselves they’re at 3.
The levels below are a diagnostic and a plan. So use them to find where you actually are. Be **honest **while reading. The coaching only works if you are.
And don’t skip ahead. That’s how companies spend six months building the wrong things.
📢 A quick word before we get into it:
The fastest way to tell what level your team is at: hand them a tool and watch
A Level 1 team uses Lovable to draft a landing page nobody ships. A Level 3 team spins up ** subagents** that research the codebase, audit the build, and synthesize a dataset in parallel, while the founder is in a meeting.
Subagents just shipped. Anton Osika put it directly: “AI products should hide complexity without hiding power.”
It is the closest thing I have seen to letting a Level 1 team operate like a Level 3 one, without rebuilding the stack.
AI Corner readers get 10% off through this link.
Now back to the diagnostic:
Table of Contents
Level 1 — The Isolated User
Level 2 — The Enthusiastic Team
Level 3 — The Systems Builder
Level 4 — The Transformed Organization
Level 5 — The Continuous Reinventor
A Final Honest Note
LEVEL 1 — The Isolated User
“I’ve been playing with AI. It’s pretty cool.”
What This Actually Looks Like
At Level 1, AI is a personal experiment. Someone on your team (or even you) has started using ChatGPT or Claude in their personal time and is now occasionally using it at work.
Usage is sporadic. There’s no shared language around it. No one is teaching anyone else.
The defining characteristic of Level 1 is that AI lives in individual browser tabs. When the tab closes, the work is gone. Nothing is saved, systematized, or repeatable. If that person goes on holiday, the AI usage of the entire team effectively drops to zero.
Diagnostic Signals — Are You Here?
If you answered yes to most of these, you are at Level 1. Welcome! Everyone starts here. But the danger isn’t being here. The danger is thinking you’re not. The Real Ceiling at Level 1
The ceiling at this level is isolation. Every person is rediscovering the same things independently. One person figures out a great way to summarize meeting notes. Another person is spending 45 minutes doing it manually. Neither knows about the other.
So the compounding knowledge that makes AI transformative simply cannot happen when everything lives in individual conversations that disappear.
There is also a **confidence **problem. People at Level 1 often don’t know what they don’t know. They have seen AI write a paragraph or answer a question, but they have no mental model for what AI can actually do at scale. This limits what they even think to try.
Coaching Plan: How to Move to Level 2
Step 1: Create psychological safety around experimentation.
The single biggest barrier at Level 1 is fear of looking foolish. People don’t want to admit they’re new to this.
So run a team session. Call it an “AI Show and Tell” where everyone shares something they tried, whether it worked or not. Normalize failure. The goal is to make experimentation feel like the expected behavior, not an extracurricular one.
Step 2: Pick one shared tool and commit to it.
Tool fragmentation at this stage is a distraction. Pick one primary AI interface (Claude, ChatGPT, Gemini) and make it the team standard. Reduce the cognitive overhead of “which tool do I use?” The best tool is the one your team will actually use consistently.
Step 3: Identify your early adopters and amplify them.
There are always one or two people who are naturally ahead. Find them. Give them time, recognition, and permission to experiment more. Make them informal champions. They will do more to move the team than any top-down mandate.
Step 4: Assign one concrete use case this week.
The fastest way out of Level 1 is a single forcing function: one specific task, done with AI, by everyone on the team, this week. Not “use AI more.” Something specific. “Summarize your last three client meetings using AI and share the output in Slack.” Specificity kills paralysis.
Step 5: Start a simple shared document of what works.
A Google Doc. A Notion page. A Slack channel. It doesn’t matter. Start capturing the prompts, approaches, and use cases that actually produce good results. This is the seed of your future prompt library and institutional knowledge base.
LEVEL 2 — The Enthusiastic User
“Our team is using AI. It feels magical. We’re saving time.”
What This Actually Looks Like
Level 2 is seductive. There’s genuine enthusiasm, visible results, and people feel meaningfully more efficient. Leadership is paying attention. There might even be a shared Slack channel where people post cool AI outputs. AI has moved from personal experiment to team conversation.
But here’s the truth: almost nothing structural has changed. The work is still being done the same way it always was. AI is a faster typewriter, not a new system. Headcount is the same. Processes are the same. The org chart looks identical to two years ago.
The danger at Level 2 is that the “magic feeling” creates the illusion of transformation. Teams get comfortable here. It feels like progress.
Well, it is progress, but it’s the easy part.
Diagnostic Signals — Are You Here?
If yes to most: you’re at Level 2. This is where most companies in 2026 are sitting. Comfortable. Not yet transformed. The Real Ceiling at Level 2
The ceiling here is person-dependency. Every AI workflow lives inside a person’s head or their individual chat history. If Sarah leaves, Sarah’s AI workflows leave with her. You haven’t captured institutional knowledge, you’ve just given individuals faster keyboards.
The other ceiling is scope limitation. Teams at Level 2 are using AI for tasks that were already being done. No one has asked the more important question: what work should we be doing that we haven’t been able to do before because it was too expensive or time-consuming? That’s where the real value hides.
Coaching Plan: How to Move to Level 3
Step 1: Systematize what’s already working.
Audit what your team is actually using AI for. Identify the five most common use cases.
Now codify them: write proper, reusable prompts with instructions, context, and output format specifications. Move these from individuals’ heads into a shared, versioned library. Give it a home with ownership. Someone whose job it is to maintain and improve it.
Step 2: Connect AI to at least one real data source.
The leap from Level 2 to Level 3 begins when AI stops working only on what you paste into it.
Identify one system (a CRM, an analytics dashboard, a financial tool) and find a way to pull data from it into your AI workflow.
This might be a simple export and upload process at first. That’s fine. The point is to experience what happens when AI has context it didn’t have before.
Step 3: Redesign one full workflow end-to-end. Don’t add AI to an existing process. Redesign the process around AI.
Take one high-frequency, high-effort workflow (a weekly report, a client brief, a variance analysis) and ask: if AI is doing the heavy lifting, what does this process look like from scratch?
The answer is usually radically different from what you have today.
Step 4: Measure and publicize the gains.
At Level 2, people feel the magic but can’t quantify it. Start measuring.
Time saved per task. Quality scores. Volume handled. Even rough numbers create accountability and excitement. Share them. Nothing accelerates adoption like proof.
Step 5: Hire or designate an AI translator.
You need someone who understands both the business problems and the AI capabilities well enough to bridge them.
This isn’t a data scientist or an IT person. It’s a curious, operational thinker who can look at a messy business process and reimagine it. Find them internally first. If you can’t, hire for it.
LEVEL 3 — The Systems Builder
“Our AI agents are working across tools. Real automation is happening.”
What This Actually Looks Like
Level 3 is where things get genuinely interesting, but also genuinely uncomfortable.
AI agents are now connected to multiple systems. Work is happening without direct human initiation. A trigger in one system sets off a chain of actions across others. Things are getting done overnight that used to require a Monday morning meeting.
Uncomfortable, because people start looking at their role and asking hard questions. If an agent can pull data from three systems, flag anomalies, draft a memo, and route it to the right person.. what exactly does the analyst do now?
Questions like that deserve honest answers, not reassurance.
Diagnostic Signals — Are You Here?
Level 3 is real systemic change. Most teams hit it in **pockets **where one function transforms while others stay at Level 2. That’s normal. The goal is to expand the pockets.
The Real Ceiling at Level 3
The ceiling at Level 3 is brittleness and trust.
Agents break when edge cases arise. Data is messier than anyone anticipated. Outputs need more checking than you hoped. Teams build impressive systems and then spend enormous energy maintaining them and debugging failures. This creates fatigue and skepticism.
The other ceiling is organizational resistance.
When systems start doing things automatically, people get nervous. Privacy concerns, compliance questions, “what if it’s wrong?” objections.
These are real concerns wrapped in legitimate fear. The teams that navigate this fastest are the ones that build robust feedback loops. That build ways for humans to catch, correct, and teach the system from its mistakes.
Coaching Plan: How to Move to Level 4
Step 1: Build a “trust framework” for your agents.
Define explicitly:
What decisions can agents make autonomously?What requires a human in the loop? What requires human approval before action?
Create a tiered model. Low-stakes, reversible actions get full autonomy. High-stakes, irreversible actions always get human review. That’s what makes agents trustworthy enough to actually use at scale.
Step 2: Instrument everything.
If you can’t see what your agents are doing, you can’t trust them, improve them, or explain them to your leadership and regulators. Build logging, audit trails, and monitoring into every agent workflow from day one. This is the boring infrastructure work that separates serious AI programs from impressive demos that never scale.
Step 3: Create feedback loops into your agents.
The best AI systems get smarter over time because humans correct them and those corrections get incorporated.
Build a simple mechanism (a Slack reaction, a thumbs down button, a weekly review) where humans can flag bad outputs. Someone should own the process of periodically reviewing these flags and updating the system’s instructions, context, or logic accordingly.
Step 4: Now ask the hard organizational question.
With agents handling meaningful workflow volume, sit down with your leadership team and ask: if we doubled our output with the same headcount, what would we do with it? And if some roles are genuinely redundant, what’s the right way to handle that?
Level 3 forces a talent and org design conversation. Don’t avoid it. Lead it.
Step 5: Expand successful patterns across functions.
Find the one or two agent workflows that are working exceptionally well. Document the pattern: the data sources, the trigger, the decision logic, the output format, the human touchpoints.
Now find three other processes that fit the same pattern. Clone and adapt. Scaling at Level 3 is about pattern replication, not building every workflow from scratch.
LEVEL 4 — The Transformed Organization
“We’ve restructured around AI. The org chart has changed. The math has changed.”
What This Actually Looks Like
Level 4 is where AI stops being a tool the organization uses and starts being a force that reshapes the organization itself.
Reporting structures change. Roles merge. Headcount grows more slowly than revenue. Teams that used to have five people accomplish the same or more with two, because the other three were primarily doing coordination, formatting, routing, and basic analysis; all things agents now handle.
This is also where the best talent starts to self-select in. People who want to be at the frontier of how work gets done are drawn to Level 4 companies. People who want certainty and stability are not. Knowingly or not, your culture has made a choice.
Diagnostic Signals — Are You Here?
Very few companies are genuinely here yet. If you are, you have a significant and compounding advantage.
The Real Ceiling at Level 4
Paradoxically, the ceiling at Level 4 is speed of reinvention.
Organizations that restructure successfully can become attached to the structure they built. They optimize the new org design relentlessly, and then a new model capability arrives that makes part of that design obsolete again.
The teams that get stuck are the ones that treat Level 4 as a destination rather than a waypoint.
The other risk is capability concentration. When teams get very small and very AI-dependent, key-person risk becomes acute in a new way.
If the one person who understands the agent stack leaves, you could lose capabilities that took 18 months to build. Documentation and knowledge transfer become mission-critical disciplines. Coaching Plan: How to Move to Level 5
Step 1: Build a permanent “reinvention rhythm.”
Schedule a quarterly process where each team asks:
*What have we built in the last 90 days that we would build differently today?*What new capabilities exist that we haven’t explored?
This exercise creates discipline. The teams that reach Level 5 treat reinvention as a normal operating cadence, not an emergency response.
Step 2: Invest heavily in documentation as infrastructure.
At Level 4, your AI workflows are core business infrastructure. Treat them like it. Every agent, every prompt system, every integration should be documented as thoroughly as your financial controls.
This is the thing that keeps your capabilities alive when people move on.
Step 3: Start thinking about AI as a competitive intelligence function.
At Level 4, you are sophisticated enough to start systematically monitoring what AI can newly do and mapping it against your processes.
Assign someone (or an agent) to track major model releases and capability updates, and to translate them into “here’s what this unlocks for us specifically.” This turns AI advancement from a distraction into a strategic signal.
Step 4: Hire for reinvention tolerance, above almost everything else.
The people who thrive at Level 5 are not the best prompt engineers or the best coders. They are the people who genuinely enjoy building something, seeing it become obsolete, and building something better.
This is a personality trait more than a skill. It’s rare. Hire it aggressively when you find it.
LEVEL 5 — The Continuous Reinventor
“We rebuild before we have to.”
What This Actually Looks Like
Level 5 is a posture. A culture. A genuine organizational belief that the half-life of any current process is short, and that this is exciting rather than exhausting.
At Level 5, the question “how do we do this?” is always followed by “and how might we do it completely differently in six months?” Teams here are not chasing AI — they are ahead of it, in the sense that they have built the muscle to adapt before the adaptation is forced on them.
There is no fixed org design at Level 5. No fixed workflow. No sacred process. The only thing that is fixed is the commitment to the customer outcome and the business objective. How you get there is always in question.
Diagnostic Signals — Are You Here?
Honestly? If you’re asking whether you’re here, you’re probably not. Level 5 organizations don’t agonize over their level. They’re too busy building.
Coaching Plan: How to Stay Here
There is no “staying here” without continuous motion.
Level 5 is defined by reinvention, which means any period of consolidation is the beginning of regression. The coaching plan at this level is more philosophical than tactical.
Step 1: Protect the builders.
The people who actually build your AI capabilities are your most strategically important employees. Protect their time. Protect them from bureaucracy. Pay them accordingly.
Step 2: Distinguish between healthy reinvention and chasing shiny objects.
Not every new model capability needs to be integrated immediately. Level 5 requires judgment. The ability to distinguish between a new capability that is genuinely transformative for your specific business and one that is interesting but peripheral.
That judgment is a leadership skill, and it needs to be cultivated deliberately.
Step 3: Teach the outside world what you’ve learned.
The organizations that sustain Level 5 over time tend to be generative. They share what they’re learning through content, recruiting, and community. This creates a talent magnet and a knowledge feedback loop that compounds their advantage further.
Step 4: Never let AI literacy become a senior-only skill.
The most dangerous Level 5 failure mode is when AI capability concentrates at the top or in a specialist team.
The organizations that sustain this level are the ones where the most junior person on the team is also building, experimenting, and contributing to the collective knowledge.
A Final Honest Note
Most companies will spend 2026 convincing themselves they are at Level 3 while operating at Level 1 or 2.
A wrong self-assessment is real and costly.
The diagnostic above is designed to make that gap visible. Use it in a room full of people. Argue about it. The disagreement is usually more valuable than the conclusion.
What level you’re at today matters far less than the rate at which you’re moving. A Level 1 team moving fast is more valuable than a Level 3 team that has stopped questioning whether there’s a better way.
The only real mistake is standing still and calling it stability.