← Home

How well do your teams use AI agents, not how much?

Answer 8 questions to see where your AI value is lost, and the one step that fixes the most.

The Index measures and improves three things: how agent output is verified, how its effect is measured, and whether it is reused.

Start the assessment

Then start with the one that describes you, or see what you get at the end

The Agentic Maturity Index scores your engineering organisation on eight areas and four levels, from ad hoc prompting to agentic orchestration, and shows where your repositories sit. Free. Your result appears on screen straight away, and we email you a copy.

Jakub Wachocki

Chief Executive Officer

Corné van Dijk

Chief Solutions Officer

We read every result.

  1. 1Ad hocNo rules or context. AI use is unstructured.
  2. 2Grounded promptingRules and project context that an AI tool reads.
  3. 3Agent-augmentedAlso commands, skills and named subagents the agent runs.
  4. 4Agentic orchestrationAlso workflows that coordinate several agents together.
A repository sits at the highest level whose artifacts it contains. Your repositories draw themselves as you answer.

What you get at the end

A result you can act on, the moment you finish. This is an example; yours is drawn from your answers.

Example result

Level2

Grounded prompting

Index 38 of 100

Your answers suggest the prompt shop: AI in daily use, person by person, with a shared starting point still to build.

ContextSkillsWorkflowsGuardrailsMeasuresSpendSpreadReuse

Where your repositories sit

  • 1 Ad hoc
  • 2 Grounded prompting
  • 3 Agent-augmented
  • 4 Agentic orchestration

Your next step

Measure output and rework from the code against a pre-AI baseline, before adding more agents.

  • Your level and pattern

    Where you stand from ad hoc to agentic orchestration, and which of seven patterns your answers fit.

  • Your repositories on the ladder

    How your repositories spread across rules and context, skills and subagents, and multi-agent workflows, from your own estimates.

  • Your profile across eight areas

    In three tracks, each with the next level and a first move you can make without us.

  • One next step

    The single step that fits your answers, a downloadable PDF roadmap if you want one, starting with where your code stands, and a link to keep your result.

Eight areas, three tracks

The first track is what your repositories contain. The second is what keeps AI output safe, measured and paid for on purpose. The third is whether the practice spreads and compounds.

In the repository

What your repositories contain that makes AI tools and agents work well.

  • Rules and context
  • Commands, skills and subagents
  • Multi-agent workflows

Control and evidence

What keeps AI output safe, measured and paid for on purpose.

  • Verification and authoritycritical
  • Measurement and trackingcritical
  • Spend and tooling

Scale and reuse

How far the practice reaches across teams, and whether it compounds.

  • Spread across teams
  • Reuse and ownership

How the score works

Four levels, each a step up from the last, following the ladder used in Stanford-based engineering productivity research: a repository sits at the highest level whose artifacts it actually contains, so level is about how AI is used, not how much. Your organisation’s level uses the same steps, and the top one also needs results that are measured, owned and reused. Every question asks what is true today, because self-assessment flatters: people who felt faster with AI were measured slower.

  1. 0 to 24

    1Ad hoc

    AI tools are used person by person. No shared rules or context, and results depend on who is prompting.

    What you getIndividuals get faster; the organisation does not learn.

  2. 25 to 49

    2Grounded prompting

    Repositories carry rules and project context that AI tools read, so the tool starts from how your project works.

    What you getMore consistent output, and a shared starting point.

  3. 50 to 74

    3Agent-augmented

    Repositories also carry commands, skills and named subagents that agents run for repeated work.

    What you getRepeated tasks run the same way every time, by whoever starts them.

  4. 75 to 100

    4Agentic orchestration

    Workflows coordinate several agents, with checkpoints and approval points. Results are measured against a baseline, and what agents build is owned and reused.

    What you getDefined work moves from ticket to reviewed change with agents doing the routine steps, and each cycle is cheaper or safer than the last.

In each row you pick one of four blocks, from ad hoc to most mature. A block counts as the middle of its quarter of the range, and an area’s score is the average of its rows. The Index weights the areas: verification and measurement 15% each; multi-agent workflows 14%; commands and skills, spread and reuse 12% each; rules and context, and spend 10% each. Your overall level can be at most one above your weakest critical area. “Don’t know” scores zero and is shown as a visibility gap. Weights are reasoned defaults, to be refitted once 200 answers are in. The repository ladder is an estimate from your answers, not a measurement of the code; the engineering benchmark reads it from the repositories themselves.

What happens to your answers

Nothing you would not expect.

Your result is worked out in your browser

It is shown on screen the moment you finish. Your private link holds the answers in the address itself.

Your details go to our CRM, with a summary

Your name, work email and company and a summary of your result (level, score and pattern) go to our CRM; every answer and score is kept in our own database. We use them to email you your result and to follow up. Our privacy policy says how we use them.

No invented peers

There is no peer benchmark yet and we will not build one from guesses. You compare with the level descriptions.

Anonymous usage counts

We count which screens people reach and how long each takes, with a random number for the visit and no name, email, cookie or IP address, so we can make the Index shorter and clearer.

Questions about the Index

How it works, who it is for, and what we do with your answers.

What is the Agentic Maturity Index?

A free self-assessment of how well an engineering organisation uses AI agents, not how much. It scores eight areas on the four levels that Stanford-based engineering productivity research uses for AI maturity: ad hoc, grounded prompting, agent-augmented and agentic orchestration. It also shows how your repositories spread across those four levels.

What do the repository levels mean?

A repository sits at the highest level whose artifacts it actually contains. Level 2 has rules and project context an AI tool reads. Level 3 also has commands, skills and named subagents the agent runs. Level 4 also has workflows that coordinate several agents. A low level does not mean low AI use: level is about how AI is used, not how much.

How is this different from the Orchestration Index?

The Orchestration Index scores how well you run delivery across suppliers, locations and AI agents. This one goes deeper on the agents: what your repositories contain, how output is checked and measured, what it costs, and whether the practice compounds. They fit together, and the next step on either can be the other.

How long does it take, and how is it scored?

There are two versions. The 3-minute check asks one row per area, eight in all, for a quick read; you can answer the rest later, with your answers carried over. In the full Index each of the eight areas has two or three rows, 17 in all, and it takes about five minutes. In each row you pick the block that describes where you are today, from the lightest (ad hoc) to the darkest (most mature). Each block scores a quarter of the range and an area’s score is the average of its rows. The overall score weights the eight areas, and your level can be at most one above your weakest critical area (verification and measurement), because AI without checks or evidence adds activity, not delivery. “Don’t know” scores zero and is shown as a visibility gap.

Is the repository ladder a measurement?

No. It is drawn from your own estimates of how many repositories carry each kind of artifact, so it shows your view, not the code. A measured ladder, read from the repositories themselves, is part of our engineering benchmark.

What happens to my answers?

Your result is worked out in your browser and shown on screen straight away. Your name, email and company and a summary of your result go to our CRM, and every answer and score to our own database, so we can email you your result and follow up. There is no peer benchmark yet, and we will not invent one.

Is it a sales tool?

It recommends one next step that fits your answers, and the first move for every area is one you can make without us.

See how well your teams use AI agents.

A 3-minute check or the full Index, about five minutes. Your result appears on screen straight away, and we email you a copy. Then one next step that fits your answers.