AI Maturity in Design Teams: Levels and How to Move Up
An AI maturity model for design maps how a team moves from personal use to real infrastructure. See the levels, how to measure them and what to do to move up.
What levels does the model describe?
The most cited AI maturity model for design started at 1Password. Its author is Matt Davey, the company’s Chief Experience Officer, and it has five levels. Jakob Nielsen summarized it and added a sixth. These are the levels as Nielsen summarizes them:
| Level | What happens in the team |
|---|---|
| 1. Limited | AI is a personal aid for administrative tasks. There is no strategy or budget, prototyping is manual and AI in the product is a superficial add-on. |
| 2. Reactive | Experiments are ad hoc and often driven by pressure to “ship AI”. Results are inconsistent and ethics is an afterthought. |
| 3. Developing | AI speeds up prototyping and communication, starting from concrete problems. There are documented playbooks and some tool investment. |
| 4. Embedded | AI is funded as a core part of the design stack, and design and engineering systems are aligned. Prototypes are built from real front-end components and designed for trust, uncertainty and refusal. |
| 5. Leading | AI is foundational infrastructure. Designers think in code and systems, and treat responsible AI as a product quality issue. |
| 6. Symbiotic (added by Nielsen) | Design time and runtime blur: AI agents generate and optimize interfaces in real time for each user. Designers define constraints and guardrails. |
The original model scores six characteristics: leadership on AI, strategy and budgeting, AI culture and talent, AI learning and enablement, AI agents and automation, and AI product design. A team can be further along in some than in others.
Nielsen says level 5 is “probably realistically the best we can hope for today (in 2026)”. He calls level 6 a stretch goal in 2026 for almost everyone, and says it may become plausible in 2027 only for organizations that already operate consistently at level 5. These are one author’s estimates, not a standard: use them as a compass, not an official rating.
How is maturity measured?
1Password proposes measuring it by outcomes rather than volume. Its article says maturity is not measured by how much a team can produce, but by whether AI helps it learn earlier, ask better questions, expose risk sooner and make clearer decisions. A team where everyone uses a chat every day can still be at a low level if no product decision has changed.
Nielsen adds another criterion: maturity is measured by a team’s normal performance, not its best days. One brilliant prototype made once with AI does not raise the level; making them this way repeatedly does.
In practice, the level shows in the team’s artifacts. You can ask yourself:
- Is there a written playbook for AI tasks?
- Do prototypes use real components from the design system?
- Does the team know when AI fits and when a person must decide?
If the answers are no, the team can be very active without being mature.
What does a team do to advance?
These practices come from 1Password’s article about its own team and from Nielsen’s summary:
- Use AI to frame problems, prototype flows, critique options and test assumptions, not only to produce screens.
- Build prototypes from real components. Questions then surface before engineering starts, while changing things is still cheap.
- Know when AI fits and when human judgment is needed. 1Password attributes this to teams at the embedded level. Writing it down in a playbook is a practical way to get there; the article does not detail how.
- Invest in AI fluency across the team. 1Password mentions an “AI champions” program as part of its broader investment in training.
- Share the model with the rest of the organization. 1Password publishes it in Figma so other teams can review it.
- Move up one level at a time. Nielsen warns that skipping levels creates “capability debt”, and that a level 2 team should aim for level 3, because level 4 and above are not yet within reach. He also suggests that focused effort can move an organization up about two levels per year.
Example
Illustrative example: a team of five designers is at level 2. Each person uses an assistant on their own, and two AI features reached the market without shared criteria. To move to level 3, the team picks a prototyping tool, writes a short playbook covering when it is used and when it isn’t, and requires every prototype to use design-system components. A few months later, to aim for level 4, it needs an annual budget, a connection between its prototypes and the engineering component repository, and a decision on who reviews trust and error cases. The last two do not depend on design alone.
Common mistakes
- Measuring maturity by the number of licenses or people using AI.
- Skipping levels, for example attempting an agent workflow without a playbook or real components.
- Leaving ethics for the end, as happens at level 2.
- Using the model to rank teams, which discourages trying new things.
- Judging the team by its best day instead of its usual work.
When should you not use the model?
When the team is very small and documenting playbooks and measuring outcomes costs more than it returns. A two-person team may learn faster by doing. It is also not useful if the organization can’t decide on tools: level 4 requires budget and alignment with engineering that design alone doesn’t control.
If the team is at level 3 and looking for criteria on which products should use AI, see AI product strategy and when to use AI in a product. To track progress with concrete indicators, product metrics for AI features help. Nielsen Norman Group describes AI in its study guide as a set of “gloves”: your hands still direct the work. That principle matters more as a team matures.