AI Developer Toolkit

Find your level — free Buy the Toolkit Buy Toolkit Buy the Toolkit Buy Toolkit Open the guides Read
Navigation menu
Buy the Toolkit Find your level — free Buy the Toolkit Open the guides Sign in

Find your level on the ladder and what holds you there.

The AI Dev Score places your workflow on Dan Shapiro’s autonomy ladder with 25 questions about how you plan, build, test and ship with coding agents. You get your level, the part of your workflow that holds it back and the one change that lifts it.

Start the AI Dev Score See a sample result

Free. About 8 minutes, no account, no card. Your score is on screen when you finish.

Sample resultInvented answers, real levels

33out of 75

Level 2 · Paired

Holds you backMaintain: Closed Loop & Incident Response, 2 of 12 points

Read the full sample

Used by engineering teams at

  • Grupa Morizon-Gratka
  • Ringier Axel Springer
  • TasteRay

Creator of wondel.ai Skills — 2.3k+ stars on GitHub (opens in a new tab)

Your result, section by section.

This is a sample with invented answers. The levels, sections, scoring and advice text are the real ones. Switch to see the Tech Lead and C-Level versions.

Pick a scorecard

Sample result AI Dev Score. Invented answers.

33out of 75

Level 2 · Paired

IDE-AI down, basics of instruction files and a few MCPs. One agent, one task, you watching every step — Level 2, where most developers are.

Points in each section

The lowest share is flagged
  1. Stage 1 Plan: Intent & Foundations 8/12
  2. Stage 2 Design: Specs & Policy Skills 4/9
  3. Stage 3 Build: Plans, Rules & Parallelism 10/18
  4. Stage 4 Test: Feedback Loops & Continuous Evals 4/12
  5. Stage 5 Deploy: Review Loops & Production Gates 5/12
  6. Stage 6 Maintain: Closed Loop & Incident Response Holds you back 2/12

The rung you're missing

Maintain: Closed Loop & Incident Response 17% · 2/12

Maintenance is completely reactive. Version a monitoring policy, schedule evidence-producing scans, and connect a scoped diagnostic agent to incident intake.

Your level is capped by your weakest area. This is it — and the fastest thing to fix.

Your full report is one step away. Enter your email to see your personalized fixes and your 30/60/90 playbook. We email you a copy too.

Sample result Tech Lead Scorecard. Invented answers.

29out of 75

Level 2 · Paired

You've started sharing practices. Some devs pair with an agent well, results still vary. Level 2 — systematize it.

Points in each section

The lowest share is flagged
  1. Shared standards & context 7/12
  2. Review & quality gates 5/12
  3. Rollout & adoption 6/15
  4. Team-scale workflow 4/12
  5. Enablement & mentoring 4/12
  6. Team ops & governance Holds your team back 3/12

The rung you're missing

Team ops & governance 25% · 3/12

Ops is ad hoc and risky. Put cost visibility, secrets policy and guardrails in place.

Your level is capped by your weakest area. This is it — and the fastest thing to fix.

Top risks to watch

  1. 1

    Underinvestment in Team ops & governance

    Section scored 25% — bottom three areas drive the next-quarter risk surface.

  2. 2

    Underinvestment in Team-scale workflow

    Section scored 33% — bottom three areas drive the next-quarter risk surface.

  3. 3

    Underinvestment in Enablement & mentoring

    Section scored 33% — bottom three areas drive the next-quarter risk surface.

Baseline wording shown; your result names your own risks.

Your full report is one step away. Enter your email to see your personalized fixes and your 30/60/90 playbook. We email you a copy too.

Sample result C-Level / Founder Scorecard. Invented answers.

41out of 75

Level 3 · Review manager

Layered review, internal MCP, metrics — agents write, engineers review, Level 3. Key move: parallelism with guardrails + an AI lead roadmap.

Points in each section

The lowest share is flagged
  1. Adoption & spend 9/12
  2. Shared infrastructure 8/12
  3. Quality gates 8/15
  4. Parallelism at team scale 6/9
  5. Org enablement Holds your org back 5/15
  6. Strategy & ROI 5/12

The rung you're missing

Org enablement 33% · 5/15

Onboarding and operating knowledge are not yet reproducible. Build one safe onboarding fixture and turn failed runs, reviews, and incidents into owned artifacts.

Your level is capped by your weakest area. This is it — and the fastest thing to fix.

Top risks to watch

  1. 1

    Underinvestment in Org enablement

    Section scored 33% — bottom three areas drive the next-quarter risk surface.

  2. 2

    Underinvestment in Strategy & ROI

    Section scored 42% — bottom three areas drive the next-quarter risk surface.

  3. 3

    Underinvestment in Quality gates

    Section scored 53% — bottom three areas drive the next-quarter risk surface.

Baseline wording shown; your result names your own risks.

Your full report is one step away. Enter your email to see your personalized fixes and your 30/60/90 playbook. We email you a copy too.

How the 75 points add up.

Every question is worth 0 to 3 points. Your total sets your level, and the section with the lowest share of its own points is flagged as what holds you back. That is arithmetic on your answers, and no AI model decides any of it.

  1. Stage 1

    Plan: Intent & Foundations

    Whether you capture intent.md before code is written, choose agent-first tools, and route modern models.

    12pts
  2. Stage 2

    Design: Specs & Policy Skills

    Whether requirements and design collapse into a single session producing spec.md guided by policy skills.

    9pts
  3. Stage 3

    Build: Plans, Rules & Parallelism

    Whether plan mode (plan.md) is the default, institutional rules are versioned, and parallel worktrees are used.

    18pts
  4. Stage 4

    Test: Feedback Loops & Continuous Evals

    Whether sessions check their own work with quantifiable targets, test-first fixing, and continuous CI evals.

    12pts
  5. Stage 5

    Deploy: Review Loops & Production Gates

    Whether layered AI review, PR babysitting, and strict human production approval gates govern releases.

    12pts
  6. Stage 6

    Maintain: Closed Loop & Incident Response

    Whether production metric breaches and security scans write the next intent.md without waiting for a human to start.

    12pts

  1. Shared standards & context

    Whether the whole team's agents start from the same rules and context.

    12pts

  2. Review & quality gates

    Whether AI output reaches review production-ready, not as slop to clean up.

    12pts

  3. Rollout & adoption

    Whether AI is actually used across the team — and whether you can prove it.

    15pts

  4. Team-scale workflow

    Whether parallelism and automation run across the team, not just one person.

    12pts

  5. Enablement & mentoring

    Whether the team keeps leveling up together as the tools change monthly.

    12pts

  6. Team ops & governance

    Whether cost, secrets, permissions and tool choice are handled at team scale.

    12pts

  1. Adoption & spend

    Whether you can see what your team is doing with AI.

    12pts

  2. Shared infrastructure

    Whether the team works on shared infra or everyone reinvents.

    12pts

  3. Quality gates

    Whether AI-authored code is gated against more bugs reaching prod.

    15pts

  4. Parallelism at team scale

    Whether you multiply engineers or just bury them in review.

    9pts

  5. Org enablement

    Whether the team learns faster than the tech changes.

    15pts

  6. Strategy & ROI

    Whether you can defend the AI budget to the board next year.

    12pts

Six sections, one total75 points

What each total means.

The four bands are 19 points wide. The scorecard places you on Levels 1 to 4 of Dan Shapiro’s ladder, because the other two cannot be tested this way.

  1. 0

    By hand

    Shapiro: Spicy autocomplete

    “not a character hits the disk without your approval”

    Not scored

  2. 1

    Assisted

    Shapiro: The coding intern

    “you offload specific, discrete tasks to your AI intern”

    0–18 points

  3. 2

    Paired

    Shapiro: The junior developer

    “where 90% of ‘AI-native’ developers are living right now”

    19–37 points

  4. 3

    Review manager

    Shapiro: The developer

    “Your life is diffs.” “almost everyone tops out here”

    38–56 points

  5. 4

    Spec manager

    Shapiro: The engineering team

    “write a spec” … “leave for 12 hours, and check to see if the tests pass”

    57–75 points

  6. 5

    Dark factory

    Shapiro: The dark software factory

    “a black box that turns specs into software”

    Not scored

Level 0 is not on the test: if you approve every character yourself, you know. Level 5 is not on it either, because a self-assessment cannot verify a dark factory. Level names and quotations: Dan Shapiro, “The Five Levels: from Spicy Autocomplete to the Dark Factory”, danshapiro.com, 2026-01-23 (opens in a new tab). Assisted, Paired, Review manager and Spec manager are the scorecard’s own labels for the four bands.

“level 2, and every level after it, feels like you are done. But you are not done”

Dan Shapiro, “The Five Levels: from Spicy Autocomplete to the Dark Factory”, 2026-01-23. This is why the questions ask what you do today.

The next eight minutes.

  1. About 8 minutes

    Answer 25 questions.

    Pick the answer that matches what you do today; one question asks you to tick every MCP server you use. Your progress is saved in this browser, so you can stop and come back.

  2. When you finish

    Read your level.

    Your score, your level and your six section scores come first, with the lowest-share section flagged. No account and no card.

  3. Optional

    Get the plan.

    Enter your email for your personalized fixes and your 30/60/90 plan, and we send you a copy. If you are signed in, this step is skipped.

Question 1 of 25 Stage 1, Plan: Intent & Foundations

How does your primary AI coding setup handle repository work?

  • 0 pts I don't use AI for coding
  • 1 pt Chat-only suggestions copied into the repository by hand
  • 2 pts An agent reads and edits the repository, but workflow and verification remain ad hoc
  • 3 pts Primary harness uses versioned instructions, plan and artifact gates, bounded tools, and reproducible verification evidence

Why it matters, as shown on the question screenClaude Code, Cursor, and Codex can all support an artifact-driven lifecycle. The useful distinction is whether the setup can inspect the repo, follow versioned policy, execute bounded work, and return verification evidence.

The question and its answers are the real ones. The points are shown here for transparency and are not shown while you answer.

What we keep: your answers and result, your browser’s user agent, and your email once you ask for the plan (your account email and name if you are signed in). The level and section scores are plain arithmetic. The written notes and the 30/60/90 plan are drafted by Google’s Gemini model from your answers and scores; signed in, your name and email are part of that request. If the model is unavailable you get the baseline plan for your level. Privacy policy

Start the AI Dev Score

The result page ends with an offer of the Toolkit for the guides on your weakest section. Take it or leave it, your plan is yours either way.

Leading a team or an organization?

Same format: 25 questions of their own, about 8 minutes, aimed at the level of the people you are responsible for.

Tech Lead Scorecard

Which level does your team ship at?

Scores your team onShared standards, review gates, rollout, team-scale workflow, enablement, and team ops. It places the team as a whole and adds your top three risks.
Start the Tech Lead Scorecard Free, about 8 minutes

C-Level Scorecard

Which level is your engineering org at?

Scores your org onAdoption and spend, shared infrastructure, quality gates, parallelism, enablement, and strategy and ROI. It is a read you can take to the board, with a 30/60/90 plan and your top three risks.
Start the C-Level Scorecard For CTOs and executives. Free, about 8 minutes

Before you start.

Something else? Write to support@developertoolkit.ai.

Can I stop halfway?

Yes. Your progress is saved in this browser and the scorecard picks up at the question you left. You can also take it again later: retaking clears your saved result and starts over.

Is the level an AI’s opinion?

No. Each answer carries 0 to 3 points, the points add up to your score, and the score falls in one of four fixed bands. The written notes and the 30/60/90 plan are drafted by an AI model from those numbers and your answers, so treat them as a first draft to check against your own repo.

Can a self-assessment be trusted?

It measures what you tell it, so the answers are written as things you can check. The top answer to question 1, shown above, names versioned instructions, plan and artifact gates and reproducible verification evidence. If you cannot point at the file, pick the answer below it. This is also why Level 5 is not scored.

Why do you want my email?

To show your personalized fixes and 30/60/90 plan, and to send you a copy. A separate tick box asks whether you also want occasional follow-up lessons. Your level, your section scores and the flagged section are on screen before you give it, so you can skip the email entirely.

What does it cost after the result?

The score and the plan are free. Each fix points to the guide that closes it. You can read 5 guides a month free, 10 with a free account, and all of them with the Toolkit, $59.99 once for 12 months. The Autonomy Ladder and Start here sections of the library are always free.

I lead a team. Which scorecard do I take?

Take the Tech Lead Scorecard to place the team and the C-Level Scorecard to place the organization. The AI Dev Score places your own workflow.

Does this page answer your questions?

One tap, no sign-in.

not at all fully

Stop guessing your level.

25 questions, about 8 minutes, and a result you can act on this week.

Start the AI Dev Score Buy the Toolkit

The score is free and needs no account.

$59.99 once for 12 months of access. Nothing renews on its own. 30-day refund on a first purchase.

Which level is your workflow at?

Restoring your scorecard…