The short version
A team's AI impact follows its culture more than its skills. In the example organization people are skilled and use AI, but they rarely share what they learn, so impact stays behind.
The fixes are mostly about people: spread the practices a few already have, agree on how the team uses AI, say where it's heading, and ask people what worries them.
In the episode
The video's chapters. Each time opens the video on YouTube at that point.
- 0:00 (opens in a new tab) Introduction
- 1:16 (opens in a new tab) Why we needed an AI Adoption Index
- 3:11 (opens in a new tab) What the AI Adoption Index measures
- 4:18 (opens in a new tab) Readiness, adoption & impact
- 5:50 (opens in a new tab) Organizational insights
- 7:38 (opens in a new tab) Why culture matters
- 13:19 (opens in a new tab) Maturity levels & knowledge sharing
- 17:02 (opens in a new tab) AI skills & tools
- 22:19 (opens in a new tab) AI adoption gaps & barriers
- 23:45 (opens in a new tab) AI as an amplifier
- 26:05 (opens in a new tab) Team-level analysis
- 28:56 (opens in a new tab) AI archetypes
- 34:30 (opens in a new tab) Team insights & actions
- 42:17 (opens in a new tab) Comparing teams
- 45:29 (opens in a new tab) Data dimension
- 47:38 (opens in a new tab) Startup team analysis
- 48:55 (opens in a new tab) Individual analysis
- 54:44 (opens in a new tab) Individual profiles
- 56:34 (opens in a new tab) Thank you!
Why they built it
Ralabs has about 170 people and, like most software companies, invests a lot in AI: training, workshops, a new way of building software, agent skills. Its engineers try Claude Code, Codex and agents, its project managers use ChatGPT, and some managers vibe-code. Some teams are far ahead, but nobody could measure how far.
Clients asked the same: how deep is this team with AI, and what stops it from using more? Without a measure, Roman says, you can't tell where to invest. Another workshop? Better tools? Different training for juniors and seniors? More knowledge sharing, so it doesn't sit with a few people? Better practices on projects?
What it measures
They didn't want another dashboard of tokens spent and pull requests opened. The survey measures the human side instead: how people use AI, the skills they have, the value they get from it and how they share what they learn. And rather than one number for the company, it gives one to each person, each team and the whole organization.
Roman puts it in three stages. Readiness: is the company ready for AI, with its data, task tracking and requirements in order? Adoption: do people really use AI in their daily work, or only pretend because leadership pushes? And impact: the value AI creates. Many companies want to measure impact first and skip what comes before it, and most AI budgets go to tools and tokens, while the value comes from people.
Technical and non-technical roles answer different questions, and the results come together in one report.
Reading an organization
The example organization has 121 responses and scores 3.0 out of 5: level 3, Practitioner, where people use AI in real work. About 15% of its people are at levels 4 and 5, and resistance is medium: most are open to AI tools.
On its maturity map, Skills, Usage and Vision are above 3, yet Impact is only 3.0. The reason, Yevhen says, is Culture, at 2.3: whenever culture sits below skills, usage and vision, impact sits below them too. Speeding up one engineer with AI is easy now. Speeding up a team takes a shared culture and shared ways of working: write ten times more code, and if your QA doesn't use AI, the team is no faster. Skills at 3.4 against Impact at 3.0 also means people can do more than they do now, so Roman would give them harder challenges.
Most people practice AI but don't share it. About half of the engineers use it only on their own, about 42% have started informal practices that nobody wrote down, and only about 5% work from shared, written guidelines. Scaling is the job: take what the 5% already do and spread it to everyone else.
The skills section shows the tools and models people use. Here most use ChatGPT in the browser, which is fine for non-technical roles, but engineers should be moving to agentic tools like Claude Code and Codex. Most use Sonnet rather than Opus, perhaps because of token limits. And most engineers follow only personal rules with sensitive data, an action point for a company that cares about security.
The report names champions, people to involve in mentoring others, and the bottom 10%, low on skills, usage and impact at once, who need a closer look. Its gaps section shows where teams need support and what blocks them: why people don't use AI more, why knowledge doesn't spread, issues with tools. Here no team has low impact, and about 40% of teams already have high usage and impact despite low culture, so better culture could take them further.
Yevhen also looks at how many people fear AI will replace them. Results like these can read as a list of who to let go, and that's not the intent: the point is how AI can amplify a team without cutting anyone. The fear itself can grow into a culture of resistance. Roman adds a line he once heard: smart people just get smarter with AI.
AI archetypes
Put ten leaders in one team and you get conflict more than results. A team needs a mix: someone who brings new ideas and someone skeptical of them, since the truth is usually in between. Yevhen sees the same with AI. Someone has to follow the news and bring in new capabilities, and someone has to question them, because much of what reaches the market is closer to a prototype than to real practice. Chase every new thing and you never get real impact. They talk through four of the archetypes:
- Skeptical Performer. Good usage, skills and impact, but low culture. Often skeptical because they know a lot and have good arguments, such as believing large language models have hit their limits. They get results but don't bring others along.
- Outcome Hunter. Gets results from new ideas, approaches and tools, on their own. The most common archetype today, Yevhen thinks: many people set up their own environment for their own productivity.
- Habit Builder. Turns their own way of working with AI into the team's. In the example, these are the people to ask to spread the 5% of written practices.
- AI Native. The goal: everyone understands the frameworks and tools and could join any team that works with AI.
The 360° has eight archetypes (opens in a new tab), for a person, a team or the whole organization.
Reading a team
The first team, AI CRM, has 22 answers. Its skills and impact are high, but culture and governance are its weakest dimensions. Three of its people are at levels 4 and 5, which is a lot, so Roman would involve them in building governance with the client: agree on a set of tools, on skills and agents to reuse, perhaps an AI code reviewer.
For culture he'd start weekly or bi-weekly sessions, ideally with the client, since a team follows a client who believes in AI and shares a vision. Leadership can explain where the team is heading, and in a small team it's worth talking with everyone to find out what they fear.
The team has one AI Native, which Roman calls lucky, but most of its people are Early Explorers at level 2, with few Visioneers. What it may miss most is a view of where it's heading, which may be why its culture is low. Sometimes it's enough to say where the team wants to be in three months.
On the usage and impact chart, the team's people show in color over the rest of the company in grey. Most are on the high-impact side, but a few are low on both: good candidates to pair with a colleague who's ahead. The report also shows people who could spread what they know but don't yet, worth bringing in through stand-ups, shared practices and regular sessions.
Most of the team say nothing blocks them, and the blocker they name most is AI skills: find which skills are missing, then train those. About 15% say there's no agreement on how to use AI, which simple agreements fix, such as which tool everyone uses, or that AI writes the tests from now on. Only about 5% know how to keep token costs down, which matters when a team pays per token, and much less on a flat subscription.
About 65% enjoy using AI, a good sign for the adoption to come. Roman enjoys his work more since AI took the small bugs off his hands. Yevhen agrees, mostly: some tasks are easier to do than to describe, but handing the routine part to AI makes the work easier.
The member list shows each person's scores. One engineer has decent skills but very low usage. For a junior that might be expected; for a senior it's worth asking why. Maybe they're limited, or have no tasks where AI helps.
Comparing teams
The educational hub team has lower skills but higher culture, and its impact is only a little lower than AI CRM's. Yevhen reads impact as roughly the average of skills, culture and usage. Train only skills, and a team of skilled engineers who don't use them well has the same impact as before: the culture of using those skills needs work too. Its people are more Individual Experts and Habit Builders, with fewer skeptics. It performs, but as a group of individuals, and its challenge is to work as a team.
Its Data score is also below the organization's average, the blue dot on its map. Data covers how information is prepared, structured and made available, and a low score may block MCP servers, automatic documentation or shared project knowledge. At level 3, key documents and knowledge are well organized; at level 4, tools reach the data through APIs, MCP, RAG or the code repositories.
The startup team is ahead on skills, usage, impact, vision and data, with more experienced people, yet its culture and governance are as low as AI CRM's. The human side of both teams has the same challenges.
Reading one person
One engineer is an Individual Expert: high skills, lower usage and low culture. They know a lot about using AI but don't use it much. Next to the 33 people they work with across their teams, they're far more skilled, but their usage is the same as everyone's, so low usage is the team's problem, not theirs. Yevhen's guess: they joined, suggested many ways to use AI, ran into the team's limited practices and lost motivation. He'd want to talk to them.
Next to engineers alone, leaving out designers, QA and DevOps, engineers here have higher culture and usage, so this person really isn't using their skills, though the organization shows it can be done. The page also suggests their next steps and sets their self-rating against how others see them: they rate themselves higher than others do.
Two more people show the range. One has low culture but uses AI more than the average engineer: the focus there isn't tools or training, but understanding why they're skeptical and whether that can change. The other is the opposite, with high culture but usage and skills below other engineers. They're open to learning, so pairing them with a skilled engineer and some training should move them quickly to level 3 or 4. Resistance, Roman says, is much harder to change than skills.
In AI Adoption Score
The survey in the video is today's 360° AI Maturity Assessment (opens in a new tab); the video still calls it the AI Adoption Index, in an earlier version of the app. Today it scores five dimensions (Usage, Skills, Impact, Culture and Vision), with Data and Governance beside them, five levels from Observer to Native and eight archetypes, in about 15 minutes per person, with a report for every person, every team and the whole organization.
Want results like these for your own teams? Book a demo, and we'll walk you through them.