Governance first, tooling second.

Adopting AI coding agents across an organisation is a governance question before it is a tooling one. Seat cost is the visible number and rarely the deciding one; what decides it is metered usage, what an agent is permitted to reach, and whether the review process absorbs the extra output it produces.

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

Pages picked for this work
5
Steps on the reading path
16
Questions in the free scorecard
25
By-role data last updated
2026-10-12

Pages: Cost governance, Governance and autonomy, MCP security, Data privacy and policy, Verified pricing and the break-even formula.

One change, two ways to work.

One adoption decision, drawn two ways: from the seat count, and from what decides the bill.

  • Written by you
  • Read by you
  • Agent and checks

From the pricing page

One number.

Seat cost is the visible number, and rarely the deciding one.

  1. Seats The price list

From the decision

Three questions decide it.

What decides it is metered usage, what an agent is permitted to reach, and whether the review process absorbs the extra output it produces.

Policy

Approval modes, sandbox, hooks

They decide what an agent can reach, so the assurance you can give is a settings review rather than a promise.

You write and read

What decides the bill

  • Metered usage
  • What an agent may reach
  • Review capacity

Judged by cost per accepted change, never by tokens per engineer.

Your decision, checks read

Evidence

Spend per team, and what was accepted

From each tool’s own telemetry and admin controls, tied to accepted work.

You read

The bill has two halves and only one is on the pricing page. Seats are fixed; metered usage is not, and the tools differ in how they meter — some include an API allowance and bill above it, others state capacity as a multiple of a base tier. Security posture becomes a configuration question with a real answer. Approval modes, sandbox policy and hooks decide what an agent can reach, which means the assurance you can give is a settings review rather than a promise. And the productivity claim should be treated as unproven until measured. This site publishes a break-even formula rather than a multiplier, because the multiplier depends on inputs only your own team has.

Drawing
One change, two ways to work.
Source
Drawn from the by-role text for CTOs, last updated 2026-10-12; cost per accepted change is from the cost governance guide.
Note
An illustration of the by-role data, not a measurement

Where it goes wrong first.

Each guide opens with the situation it exists to fix. Here are 4, in the guides’ own words, with the page that fixes them.

  1. The scene
    “Finance messages you on the first of the month: the AI tooling line doubled again, and they want to know why.”

    The page that fixes it Cost governanceGovern Claude Code, Codex, and Cursor spend: per-team telemetry, spend limits, a model and effort policy, and a monthly review tied to accepted work.

    Open Cost governance 2026-10-17

  2. The scene
    “Your agents already open pull requests overnight. Someone asks in the incident review who approved the migration that dropped a column …”

    The page that fixes it Governance and autonomyAssign agent autonomy by risk class, not confidence: guardrails during work, gates before merge and production, named approvers, and controls you rehearse.

    Open Governance and autonomy 2026-10-16

  3. The scene
    “… forty engineers have each run claude mcp add or edited config.toml for months, a shared GitHub token sits in three dotfiles, and nobody can say which servers can write to production or what one of them did last Tuesday.”

    The page that fixes it MCP securityCTO Scorecard Q8: enforce a reviewed MCP allowlist in managed config, authorize each tool, gate writes, log every call, and test with injection fixtures.

    Open MCP security 2026-09-26

  4. The scene
    “… a developer pastes customer emails into an agent to debug a slow page, another pastes a full DATABASE_URL, a third signs in with a personal account, so no company retention terms apply.”

    The page that fixes it Data privacy and policyDecide what data may reach a coding agent: four data classes, a register of approved services, vendor retention, context controls and tests that prove them.

    Open Data privacy and policy 2026-09-26

Scenes are the opening scenarios of the pages and are written as examples, not as reader reports. Quotations are trimmed. The pages are documentation pages and come with the Toolkit.

Your reading path, in the order it pays off.

From the 16 steps on the CTO path, 6 picked for the governance and spend questions: the evidence, a pilot, who owns the harness, cost per accepted change. The numbers are each step’s place in the full path, and each one says what you can show when it is done.

  1. Step 1 of 16 Free

    The State of Agentic Engineering, September 2026

    The dated third-party evidence, publisher by publisher, before any decision.

    Done whenYou can cite three sources with dates for the shift you are planning.

  2. Step 5 of 16 Toolkit

    Designing a pilot that proves something

    A pilot with a control group and a stop rule, not a demo.

    Done whenPilot design v1: teams, duration, metrics and the decision it feeds.

  3. Step 6 of 16 Toolkit

    The operating model: who owns the harness, the gates and the evals

    Decide who owns the harness, the gates and the evals across teams.

    Done whenAn operating model v1 with named owners.

  4. Step 7 of 16 Toolkit

    The economics of agent-built software: cost per accepted change

    Cost per accepted change, not cost per seat, is the number to manage.

    Done whenA cost-per-accepted-change baseline for one team.

  5. Step 8 of 16 Toolkit

    Governance and autonomy: put humans at the gates

    Autonomy is granted by risk class, with humans at the gates.

    Done whenAn autonomy register v1: which change classes may merge on evidence.

  6. Step 13 of 16 Toolkit

    AI usage cost governance

    Make spend visible and attributable before it becomes a budget surprise.

    Done whenMonthly spend per team and per accepted change on one dashboard.

Steps and “done when” lines: the CTO reading path, 16 steps and 4 further reading, last updated 2026-09-26. The ladder pages are free; every other step is marked.

Read the whole path

Where your organization sits on the ladder.

Dan Shapiro’s levels describe how much of the work you hand to an agent. Each row shows what one organization looks like at a rung the scorecard can place you on.

Not sure which row is yours?

The C-Level / Founder Scorecard asks 25 questions in about 8 minutes, scores 6 sections and gives you a 30/60/90-day plan for the next rung. Free.

Take the C-Level / Founder Scorecard

No account, no card. An email only at the end, and only if you want the full playbook.

  1. 4

    The engineering team

    “leave for 12 hours, and check to see if the tests pass”

    One organizationTeams run loops that leave. The organization assigns autonomy by risk class, reads evidence at the gates, and keeps a named person reading the code for the escalation classes.

    Where the toolkit points
  2. 3

    The developer

    “Your life is diffs.”

    One organizationEvery team’s agent opens pull requests and people read the diffs. Spend and review time rise with output.

    The valley, levels 2 and 3
  3. 2

    The junior developer

    “feels like you are done. But you are not done”

    One organizationDevelopers pair with agents on their own tools, invoices rise, and nobody owns the shared rules.

  4. 1

    The coding intern

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

    One organizationA few engineers hand agents small tasks, on personal plans, outside any policy.

Level names and quotations: Dan Shapiro, “The Five Levels: from Spicy Autocomplete to the Dark Factory”, 2026-01-23 (opens in a new tab). The ladder runs from Level 0 to Level 5 and the scorecard places you on 1 to 4. The “One organization” column is our illustration.

Your first week, planned.

  1. Day 1

    Generate your Setup Pack.

    Answer 9 questions about your stack and tools, including language, framework, package manager and test runner. Your agent runs INSTALL-PROMPT.md and checks each setting against what you have installed. About 20 minutes. The Setup Pack is included in the Toolkit.

  2. Days 2 to 5

    Hand over one real change.

    Pick one pilot team. Agree the metric definitions, set the autonomy class and the spend limit, then let the team run one change through the gates, with spec.md and plan.md accepted by a named person before the agent builds, and read the evidence it produces.

  3. By day 30

    Decide with evidence.

    You have one change done the new way. If it did not change how you work, ask for a refund before day 30.

Buy the Toolkit

The Toolkit includes the guides, the Setup Pack and all eight books, for 12 months. $59.99 once for 12 months of access. Nothing renews on its own. 30-day refund on a first purchase.

Questions about this work.

Something else? Write to support@developertoolkit.ai.

What does it cost per seat?

Claude Team Standard is $25 per seat a month, or $20 billed annually (read 2026-09-26, unchanged on 2026-10-12). ChatGPT Business is $20 per user a month billed annually, or $25 billed monthly (OpenAI, read 2026-10-12). Cursor Teams Standard is $40 per user a month, or $32 billed annually (Cursor, read 2026-10-12). Metered usage sits on top of the seat and is the half that varies; read the vendors’ pages the day you buy.

How do you limit what an agent can reach?

Through approval modes, sandbox policy and hooks, which all three tools have in some form. Codex keeps sandbox policy separate from approval, so what may run and what must be confirmed are configured independently; Claude Code has six approval modes (Manual, accept edits, plan, auto, don’t ask, bypass) with allow, ask and deny rules. For an unattended run the layer that works is an exact allowlist on a disposable runner.

What do I read instead of the diff?

The evidence the change carries: the spec delta, the acceptance results, the strength of the oracle and runtime signals from a running build. A named person still reads the code for authentication, money, schema and migrations, changes to the tests themselves, and anything no check covers.

I only use one of the three tools.

That works. Where Claude Code, Codex and Cursor differ, the guides and recipes show a tab for each, so you read the one for your tool.

Ask for evidence before you grant autonomy.

Within the 30-day refund window, scope one pilot with the metrics, the gates and the spend limit written down.

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