Review is the bottleneck. Design for it.
A team adopting AI coding agents does not have an individual problem repeated N times. Output rises faster than review capacity, and standards that lived in people's habits stop propagating — which makes shared config files, agreed approval modes and a review process that scales the three things worth settling first.
$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
One week of agent output, drawn two ways: read by everyone, and routed through a queue.
- Written by you
- Read by you
- Agent and checks
Review as a habit
Every diff, every time.
When every engineer can produce a forty-file change before lunch, the constraint moves to who reads it.
- Diff A person reads
- Diff A person reads
- Diff A person reads
- Diff A person reads
Review as a queue
Checks first, a named human last.
Layered review runs deterministic checks first, then review agents that each answer one question, then a named human.
Rules
One config file
A convention that used to spread through code review is written into the file every agent reads.
You write and read
The layered review
- Deterministic checks
- Review agents
- A named human
Each pass has its own evidence format and says what may block a merge.
Checks, agents, a named human
Evidence
The evidence bundle
What an agent’s pull request must prove, so review reads evidence.
You read
Review becomes the bottleneck, not writing. When every engineer can produce a forty-file change before lunch, the constraint moves to who reads it — and a team that does not change how it reviews will simply merge less carefully. Standards stop travelling by osmosis. A convention that used to spread through code review now has to be written into a config file every agent reads, or it will be silently reinvented by each of them.
- Drawing
- One change, two ways to work.
- Source
- Drawn from the by-role text for tech leads, last updated 2026-10-12; the layers are from the layered pull request review 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.
- The scene
“You turned on an AI reviewer last month. It posts 15 comments per pull request, most of them about naming the formatter already settles, and last week it missed an unscoped query that returned another tenant’s rows.”
The page that fixes it Layered PR reviewSet up Claude Code, Codex and Cursor review agents from one REVIEW.md: separate passes, evidence-backed findings, a CI gate, and named human sign-off.
- The scene
“… an engineer vibe-coded a pricing calculator in an afternoon to settle an argument with sales. Four months later, account managers send its link to customers, it reads a copy of the CRM, and it runs on the engineer’s personal cloud account.”
The page that fixes it Setting a vibe-coding policyCTO Scorecard Q16: a prototype policy you can adopt, with four exposure classes, automated graduation triggers, and a six-criterion gate before production.
- The scene
“Three teams run three agents, the AI line doubled, and nobody owns the shared rules or the merge gate.”
The page that fixes it Team and enterprise workflowsThe CTO's hub for coding agents at company scale: operating model, autonomy gates, security, measurement, spend and hiring, plus a baseline to adopt.
- The scene
“Finance wants one number for eight engineers before the renewal. One engineer already pays for Max 20x on a personal card, and the comparison someone forwarded quotes prices nobody can source.”
The page that fixes it Team seat mathsWhat each coding-agent plan costs on 2026-10-12, how usage is metered beyond the seat, and a worksheet that prices a team from real invoices.
Your reading path, in the order it pays off.
From the 16 steps on the tech lead path, 6 picked for review capacity and standards that travel: one map, evidence, a shared rules core and the review queue. The numbers are each step’s place in the full path, and each one says what you can show when it is done.
-
Step 1 of 16 Free
One map: the autonomy ladder, the lifecycle and the factory stations
One frame for maturity, process and capability before you change how eight people work.
Done whenYour team placed on the map, loop by loop.
-
Step 3 of 16 Free
Reading evidence instead of code
The standard the team moves to: evidence first, code only for escalation classes.
Done whenEscalation classes agreed with the team.
-
Step 8 of 16 Toolkit
Shared agent rules — one core, tested adapters
One shared, versioned rules core instead of eight private setups.
Done whenShared rules in the repository, reviewed like code.
-
Step 11 of 16 Toolkit
The evidence bundle: what an agent's pull request must prove
Define what every agent pull request must prove, in one manifest.
Done whenThe evidence bundle required by CI on agent pull requests.
-
Step 14 of 16 Toolkit
Running the review queue when agents open the pull requests
Run the review queue as a system when agents open most of the pull requests.
Done whenA work-in-progress limit and a queue-age alert on the review queue.
-
Step 15 of 16 Toolkit
Helping the team stop reading every diff
Move the team from reading every diff to trusting the evidence, deliberately.
Done whenOne risk class merged on evidence alone for a month without an incident.
The guides for this work.
These are documentation pages, and they come with the Toolkit.
Where your team sits on the ladder.
Dan Shapiro’s levels describe how much of the work you hand to an agent. Each row shows what one team looks like at a rung the scorecard can place you on.
Not sure which row is yours?
The Tech Lead 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.
No account, no card. An email only at the end, and only if you want the full playbook.
- 4
The engineering team
“leave for 12 hours, and check to see if the tests pass”
One teamThe team writes specs and reads evidence bundles. Review is a queue with limits: a named human reads the code for the escalation classes and approves the rest on its evidence.
Where the toolkit points - 3
The developer
“Your life is diffs.”
One teamEvery engineer’s agent opens pull requests and people read every diff, so review is the bottleneck.
The valley, levels 2 and 3 - 2
The junior developer
“feels like you are done. But you are not done”
One teamEngineers pair with agents one by one, each their own way, and standards travel by habit.
- 1
The coding intern
“you offload specific, discrete tasks to your AI intern”
One teamA few engineers hand agents small tasks, and the team’s process does not change.
Your first week, planned.
-
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.mdand checks each setting against what you have installed. About 20 minutes. The Setup Pack is included in the Toolkit. -
Days 2 to 5
Hand over one real change.
Pick one repository and one loop. Write the shared rules file, then
intent.mdandspec.mdfor one change, have the agent draftplan.mdand accept it before any code changes, let the agent build, and send the result through the layers. -
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.
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.
Which tool to use
Questions about this work.
Something else? Write to support@developertoolkit.ai.
What does a team seat cost?
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). The pricing analysis verifies no included-usage amount for Cursor’s plans, so compare included usage as well as seat price, and read the vendors’ pages the day you buy.
How do you keep standards consistent across a team?
Put them where every agent reads them: CLAUDE.md, AGENTS.md or .cursor/rules, committed to the repository. Anything that must happen every time belongs in a hook rather than in a convention people are expected to remember.
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.
Move the team up one rung, not one person.
Within the 30-day refund window, try one loop in one repository with the shared rules and the layered review.
$59.99 once for 12 months of access. Nothing renews on its own. 30-day refund on a first purchase.