Make the whole company AI‑native. Engineering goes first.

Machines do the hands-on work. People write the intent, make the calls and take the escalations. This is the playbook for the CTO who has to get every function there, one gate at a time, with every excuse on trial.

Take the challenge →Read the guide

5 rungs · 5 gates · 10 excuses on trial · 10 chapters · 87 min read · 0 promised durations

A typical first reading. Yours goes here after intake, and it is rarely flattering.

Who touches the keyboard?

Rung 2: the interpreter

  1. Ask
  2. Person prompts
  3. Model answers
  4. Person pastes
  5. Output

The person pays a tax in attention on every exchange, and the work stops when they do.

Rung 5: the intent-setter

  1. Intent, written once
  2. Machine does the work
  3. Output

The person writes what matters once, then takes only what the intent does not cover.

Pick any workflow. Watch who produces the output. If it is a person, that workflow is not AI-native yet, however much AI they paste along the way.

Most companies stall right there: a person in the middle, prompting, pasting, re-explaining, paying a tax in attention on every exchange. The guide exists to take that person out of the middle and put them where they belong: writing the intent and making the calls.

Five rungs. Every function climbs alone.

The rung is set by where the human sits. No skipping. A function only counts as rung 5 when every one of its core workflows is.

The ladder, in full →

  1. 1Hands on keyboard

    Human
    Does the work by hand
    Machine
    Nothing, or autocomplete
    To climb
    Tools approved and in daily use

  2. 2Interpreter

    Human
    Prompts and pastes, pays the "prompt tax"
    Machine
    Answers when asked
    To climb
    Work moves out of chat and into pipelines

  3. 3Reviewer

    Human
    Reviews every output (review is the bottleneck)
    Machine
    Drafts the work, waits for approval
    To climb
    Evals the team trusts enough to stop checking everything

  4. 4Sampled review

    Human
    Reviews a sample plus every escalation
    Machine
    Does the work, flags what it's unsure of
    To climb
    Intent written down, owned, reviewed on a cadence

  5. 5Intent-setter

    Human
    Writes intent, makes the calls, handles escalations
    Machine
    Does the work end to end, escalates when intent runs out
    The top
    Nothing above it. Hold it.

Five gates. Nobody skips one.

Engineering proves it first. One willing business function follows. Then the rest, in waves, sized to how much your people can review. Each gate is a condition, never a date.

  1. 0

    Diagnose

    Intake, place every function on the ladder, open the Constraint Court docket. The honest timeline goes to the CEO and board now, not in the last month.

    Gate

    Deadline verdict accepted, an owner named for each function

  2. 1

    Engineering proves it

    Engineering climbs to rung 3, then takes one workflow to rung 5. Sandboxes, evals and review capacity get built, and the first intent document gets written.

    Gate

    Engineering at rung 3. A willing business function can start from here.

  3. 2

    First export

    One business function with a willing owner. The CTO co-writes its intent document. The pattern is retold in that function's language, not engineering's.

    Gate

    Second function at rung 3, with its intent owned and reviewed

  4. 3

    Spread

    The remaining functions move in waves, one owner each, sized to review capacity. Constraint Court rulings from earlier phases become the playbook.

    Gate

    Core workflows at rung 5 in every function in scope

  5. 4

    Run it

    Drift reviews, escalation paths, the trust log. The leader's job becomes writing intent and handling escalations.

    No gate

    It never ends. This is the job now.

You will hear 10 excuses. Rule on them first.

Every constraint your exec team states goes to Constraint Court and comes back real, negotiable or false. 5 of these 10 are false. See how many you catch before the Court does.

0/10 Rule on each one before you read the Court's verdict.

  1. 01“Our codebase is too legacy for agents”

    Your ruling

  2. 02“We need to pay down tech debt first”

    Your ruling

  3. 03“Our seniors won't adopt it”

    Your ruling

  4. 04“We need to hire AI engineers first”

    Your ruling

  5. 05“Let's wait for the models to get better”

    Your ruling

  6. 06“Security won't approve it”

    Your ruling

  7. 07“We can't prove the ROI”

    Your ruling

  8. 08“The platform team has to build it first”

    Your ruling

  9. 09“Customer contracts forbid sending their data to a model”

    Your ruling

  10. 10“We're regulated (SOX, HIPAA, data residency)”

    Your ruling

How a hearing runs, and why a constraint nobody owns is false by default: Constraint Court →

The board said ninety days.

The guide never publishes a duration. It runs your deadline against five rules and your own numbers, and tells you whether it fits, is tight, or won't hold. Then it tells you what to say, and when.

  1. 01One rung at a time
  2. 02Engineering goes first
  3. 03Rung 5 needs one full drift cycle
  4. 04Review capacity is the ceiling
  5. 05Run functions side by side only if they have separate owners

200 engineers, SOX, a board mandate of "AI-native in 90 days"

Won't hold

What it can hold: engineering at rung 3 and one intent document in each function that faces the audit.

Say that to the board in week one, not week eleven.

The timeline reality check →

The guide

Part I · The bar

  1. 01This one is yoursEvery exec will want to own the AI transition. Only one already runs a function on pipelines. This is the CTO's to lead, at any stage, and nobody else can.
  2. 02Who touches the keyboard?AI-native is not how much AI you use. It is who touches the keyboard. If a person still produces the output, the workflow isn't there, however clever the prompt.
  3. 03Five rungsFive rungs, set by where the human sits. Score every function on its weakest workflow, climb one rung at a time, and don't call anything rung 5 until all of it is.

Part II · The run

  1. 04Phase 0: DiagnoseNothing gets built in Phase 0. You find out where every function really is, put every excuse on trial, name one owner per function and tell the board the truth.
  2. 05Phase 1: Engineering proves itYour team goes first, and the whole company is watching. Get engineering to rung 3, take one workflow all the way to rung 5, and make it impossible to argue with.
  3. 06Phase 2: First exportProving it in engineering was the easy part. Now take it to a team that has never seen a pipeline, in their words, with an owner who wants it.
  4. 07Phase 3: SpreadEngineering proved it and one function volunteered. Now the rest of the company moves, in waves, and you say out loud what happens to the jobs.
  5. 08Phase 4: Run itThere is no gate here and no finish line. Rung 5 decays the day nobody reviews the intent. Your job is now writing it and taking the calls it can't make.

Part III · The fight

  1. 09Constraint CourtEvery reason you can't go AI-native gets a hearing and a verdict: real, negotiable or false. Half the list so far is false. Bring an owner, or lose the case.
  2. 10The timeline reality checkYour deadline gets a verdict: fits, tight or won't hold. No durations, only five rules and your numbers. Then you tell the board in week one, not week eleven.

Your move.

Sign in and the guide becomes yours. It places every function on the ladder, checks your deadline against the rules, hears your constraints, and holds you to the rulings at every check-in.

Start your intake →Take a constraint to court
  1. 1Intake. Your headcount, your deadline, who set it, and the functions in scope.
  2. 2Placement. Every function on the ladder, workflow by workflow. Usually lower than you hoped.
  3. 3The Court. Every constraint gets a ruling, an owner and a date.
  4. 4The takeaways. A one-page board summary, intent templates, and the Court transcript.
  5. The same profile lives in the tinycto MCP server, so a ruling you get in Claude shows up here.