You still read every line the agent writes.That’s Level 3. There’s a Level 4.

You could, and it would sound right. Ask one about Codex approval modes and there is a good chance you get `--approval-mode`, which is not a flag — the real one is `--ask-for-approval`, and its values are `untrusted`, `on-request` and `never`. Copy the invented one and Codex rejects the command outright — you do not get a wrong permission posture, you get an agent that will not start. What you are paying for is that this list of invented flags is checked against a named release, with the version written next to each entry — and that the guides show when they were last updated. It matters more the higher you run: an agent you watch at Level 2 gets corrected in a minute; an agent left unattended for twelve hours on an invented flag is twelve hours of nothing, discovered in the morning. We keep a list of the ones models keep inventing.

At scale they already write most of it. Anthropic reports that more than 80% of the lines merged into its codebase in May 2026 were written by Claude, and Stripe merges over 1,300 pull requests a week with no human-written code (Stripe, February 2026). Across the industry the share is lower: DX measured 51.9% AI-authored code on average across more than 400 companies in Q2 2026. The other half of the evidence matters just as much. in METR’s controlled study, experienced developers’ tasks took 19% longer with AI on their own repositories (2025), and Faros AI measured incidents per PR up 242.7% alongside the throughput gains (April 2026). Which set your team gets depends on what the agent has to prove before merge — that is what the levels measure.

You own intent, architecture, acceptance criteria, risk, production approval and exception handling. Agents can research, propose, implement, test and prepare evidence, but accountability does not move to the model. The workflow makes those human decisions explicit instead of hiding them inside a chat.

That is the most common reason people cancel, so start with one real change. Use the Setup Pack to add repository context, write intent.md, spec.md and plan.md for something you were going to build anyway, and run it in your tool. You get one change done the new way before you read anything else. If it still is not for you, the 30-day first-purchase refund covers it.

Tool fluency is Level 2 or 3, not a level of trust. Ask whether every request becomes a committed intent, spec and plan; whether the agent runs isolated; whether a PR carries its own evidence; whether incidents update the next plan. Where the answer is no, that is the guide to read — for the tool you run, at the release you run.

The levels are Dan Shapiro’s Five Levels (January 2026): from Level 0, where you write everything, to Level 5, the dark factory, where agents pull work behind policy and CI. Level 2 is one agent, one bounded task, with you watching. Level 3 is agents implementing plans while you review every diff. Level 4 is you owning the spec and the gates while agents run unattended. The six stages — Plan, Design, Build, Test, Deploy, Maintain — are what you do at each level: the scorecard asks about them to place you, and the guides show the stage practice that moves you up one rung.

A tutorial usually teaches one feature in one tool, as of the day it was recorded. Here the guides sit on the ladder, cover all three tools, and each shows when it was last updated — so you can see how old what you are reading is.

Yes. From your account page, and access runs to the end of the period you paid for.

Yes — 30 days from your first purchase, for any reason. Email support@developertoolkit.ai and we refund it; we won't argue with you about why. Applies to first-time purchases, as set out in the Terms.

You keep full access until the end of the period you've already paid for — the subscription simply doesn't renew. Your account stays in place, so you can pick up where you left off anytime. Any ebooks you've downloaded are yours to keep.

Yes. Payments are handled by Polar, our merchant of record, which issues an invoice for every charge with the applicable VAT or sales tax for your country. It lands in your inbox right after payment — ready to expense or pass to your accountant.

Unlimited seats for one flat price. Everyone with an email address on your company domain gets full access automatically, and teams spread across domains can add named addresses instead — no per-seat billing, no invites to manage. Need white-label or on-premise? Ask on an Enterprise intro call.

The 100+ recipes cover React, Next.js, Vue, Python, Node.js, Go and Rust, and the ladder itself does not care what language you write — the brief chain, the verification gates and the MCP setup are the same in every stack. The recipes make it concrete in yours.