How does your engineering team work with AI agents?
See which level of AI engineering your team is at, from autocomplete to agents that run on their own, and where you sit on the leader matrix next to other engineering teams. Your score across five areas appears straight away, with a concrete next step wherever you're behind.
38 questions, about eight minutes. Your result appears on screen as soon as you finish.
Your AI engineering level
Which of the six levels from our whitepaper How AI changed engineering your team is at, and what the next level asks of you.
Your place on the leader matrix
Where your team sits against other teams: how far along the journey you are, and how well the way of working has caught up.
A next step for each gap
For every question where your team is behind, one specific change to make, with a link to how it works.
The AI engineering leader matrix
Every team that takes the benchmark becomes one anonymous dot. Across: how far the team has adopted AI tools, from how engineers work with AI to how long and how much. Up: how well the engineering practice around those tools has caught up, which is your benchmark score. The lines sit at the median of your peers, so the quadrant you land in says how you compare with teams like yours.
- LeadersAhead of peers on adoption, with review, guardrails and hiring that kept up.
- Tool-firstAhead of peers on adoption, but the practice around the tools lags behind.
- Practice-firstStronger practice than peers, with less adoption so far.
- Starting outBehind peers on both, which is where the quickest gains are.
The dots appear once twenty teams have answered. Until then your result page shows where your own team lands.
Teams appear here once twenty have answered.
Which level is your team at?
The levels follow the six waves of AI tools in our whitepaper How AI changed engineering. Your answers place your team on one of them.
- 01
Completion
AI finishes the line you're typing. The engineer still writes the code.
- 02
Chat
Engineers describe a problem and paste the answer in. The model can't see your system.
- 03
Context
The model reads your repository and a context file with your conventions, kept up to date.
- 04
Harness
Agents edit, test and fix on their own, inside a shared setup with guardrails on what they can reach.
- 05
Loop
Agents start from tickets, schedules or CI and open pull requests without someone approving every step.
- 06
Verification
The spec and the tests exist before the agent starts, and review depth follows what a change can break.
Your score: five areas, three questions each
Every area is scored out of 100 and counts equally towards your total. The other questions describe your team, so we can compare you with teams like yours.
- 01
Ways of working
How tasks reach an agent, where it finds your conventions and whether the team shares its setup.
- 02
Verification
How agent-written changes are reviewed, tested and explained before they reach production.
- 03
Platform and guardrails
Where agents run, how changes are previewed and how your code reaches AI models.
- 04
Measurement
Whether you can see what AI changes in delivery, cost and quality.
- 05
Team and hiring
How you assess AI ability when hiring and how the current team learns to work with agents.
The tools are in use, but the way of working around them hasn't been designed yet.
Parts of the team work this way, but little of it is written down or enforced.
The practice is standard across the team, and review and hiring have caught up.
The team treats its agent setup, guardrails and measurement as part of the product.
What we store
Your answers, your first name and your business email. We don't ask for your company name and we don't store your IP address.
How results are published
Only as aggregates. A group of fewer than five teams is never shown, on your result page or in the report.
Who sees your answers
Miyagami only. We don't sell your data or share your answers with anyone else. Email info@miyagami.com to have your submission deleted.
The benchmark runs on Cloudflare's network. Your answers stay with Miyagami.
How long does it take?
About eight minutes. There are 38 short questions, one per screen, and almost all take one click.
Who is it for?
CTOs, heads of engineering, engineering managers and tech leads of teams that build software, from five engineers to several hundred.
How is the score calculated?
Fifteen of the questions are scored from 0 to 3 across five areas of three questions each. Each area is scored out of 100 and your total is the average of the five, so every area counts equally. The other questions describe your team and are used for comparisons and the report.
How is my AI engineering level determined?
From the six levels in our whitepaper How AI changed engineering: completion, chat, context, harness, loop and verification. You reach a level when your answers show its practice and every level before it, for example a maintained context file for Context, or tests written before the code for Verification.
Are companies named on the leader matrix?
No. Every dot is anonymous, and the matrix only fills in once at least twenty teams have answered. Your result page shows your own dot highlighted next to the others.
Who do you compare me with?
Teams with a similar number of engineers. A comparison only appears once at least five of those teams have answered; until then we compare you with all respondents, or show nothing.
When is the report published?
In January 2027. Tick the box at the end and we'll send it to you.