# AI-DLC adoption checklist

From "SDLC vs AI-DLC: why technology leaders are switching" by Sunil Kumar
https://sunilkumar.ai/playbooks/sdlc-vs-ai-dlc/
Licence: CC BY 4.0. Copy it, edit it, use it in your own programme.

The rule that makes the rest work: the agent may propose anything and produce anything, but only a human accepts.

## Step 1. Readiness (days 1 to 15)

- [ ] CI time per change measured, with a target you will hold it to
- [ ] Test environment fidelity documented: what differs from production, who closes the gap
- [ ] Test coverage measured, and the assertions checked for meaning
- [ ] Security scanning (static analysis, dependencies, secrets) in the pipeline, or scheduled before the first unit ships

## Step 2. The first programme

- [ ] One bounded programme with an outcome written in a sentence
- [ ] Five senior people, curious rather than available
- [ ] A domain reviewer from the business side, part-time, named
- [ ] Baseline recorded for lead time, change failure rate and cost per outcome

## Step 3. The context repository

- [ ] Domain glossary in plain language
- [ ] Architecture decisions, with the reasons
- [ ] Non-negotiables: data residency, dependencies, patterns you will not accept
- [ ] Refusal list, enforced in code where possible, not only in a prompt
- [ ] Versioned in the codebase; read by every agent, updated by humans

## Step 4. The four gates

| Gate | Question it answers | Owner |
|---|---|---|
| 1. Plan | Is this the right unit of work, and is done defined? | |
| 2. Design | Is this the design we would defend? | |
| 3. Review of intent | Does the unit do what was asked, and nothing else? | |
| 4. Release | Are the artefacts here to run it, watch it and roll it back? | |

- [ ] Every gate has a named owner on your side, not the vendor's
- [ ] The acceptance rule is written down and visible to the team

## Step 5. First units (days 16 to 45)

- [ ] First elaboration session held with the whole pod and the domain reviewer
- [ ] Units small enough to review in twenty minutes
- [ ] First three units in production, each accepted by a human at gate 3
- [ ] What the agent consistently got wrong recorded in the context repository

## Step 6. Unit flow and measurement (days 46 to 75)

- [ ] Unit-flow board replacing sprint planning and grooming
- [ ] Lead time and change failure rate reported weekly against baseline
- [ ] Cost per outcome reported monthly, including model and tooling spend
- [ ] Security checks at the construction gate, with the agent fixing its own findings first

## Step 7. The decision (days 76 to 90)

- [ ] Three metrics against baseline, ROI model populated over three programmes
- [ ] Go or no-go decided, next team named
- [ ] One engineer from the pilot pod moving with the process

## ROI model (fill in)

```
Baseline (SDLC)
  Team size            ____ people
  Fully loaded cost    ____ per person-month
  Duration             ____ months
  Baseline cost        ____

AI-DLC pod
  Team size            ____ senior people
  Duration             ____ months
  People cost          ____
  Model and tooling    ____ (10 to 15 percent of people cost)
  Enablement, one-off  ____ (platform, CI, gates)
  Pod cost             ____

Direct saving          ____
Time to value          ____ months earlier
Second-order value     ____ (earlier revenue, less regulatory exposure)
```

Run it over three programmes. The first carries the enablement cost; the second and third do not.