The situation
Licences on day one. No story on day ninety.
Most AI rollouts start the same way. Licences go out, a chat channel opens, and a few keen engineers share what works. For a while it feels like progress.
Ninety days later, adoption is up and someone asks what changed. Which work went faster? What did it cost? Which decisions did people still make, and which did the tools make for them? Nobody can say. Not because nothing happened, but because nothing was set up to show it.
A rollout needs a shape: a small start, a second team that works differently, and a view leadership can read. Here is one that fits in a quarter.
The plan
Three phases, each with something to show.
Each phase has three parts: what you do, what you decide, and what you should be able to show when it ends. Don't move on until you can show it.
- 1Days 1–30
One team, one real piece of work
- Do
- Pick one team and one roadmap item that matters but isn't on fire. Run one cycle end to end: a plan, the work, a review.
- Decide
- Sign-in through your identity provider. The rules that must hold from the first day. That a person approves the plan and accepts the work.
- Show
- One item shipped, with the record of who asked, who approved and what it used.
- 2Days 31–60
A second team, working its own way
- Do
- Add a team on a different AI tool and rhythm. Share the roadmap and the rules with the first team. Hold the first retro.
- Decide
- What the second team may change for itself, and what stays the same for everyone. Which agent a team builds or adapts for itself.
- Show
- Two teams on one roadmap, and one agent improved from what the retro found.
- 3Days 61–90
A view leadership can use
- Do
- Bring in a third team. Estimate each piece of work before it's approved.
- Decide
- The four things leadership will see each quarter: what shipped, what it used, how often it was right first time, and what it asked of people.
- Show
- One page a leadership meeting can read, built from the record rather than from memory.
Before day one
Five questions to settle before day one.
Most of these take a conversation, not a project. Settle them with the leads of your first two teams.
Which team goes first, and why?
TeamPick the team with a real backlog and a lead who wants to try it, not the team with the most spare time.
Which piece of work is real enough to matter but safe enough to learn on?
WorkSomething on the roadmap that people will notice shipping, but not the change that keeps you up at night.
Which decisions must stay with a person?
DecisionsWrite them down before the first plan runs, so an agent never has to guess where the line is.
What will you measure from day one?
MeasuresAnything you don't record from the start, you'll be estimating from memory at the end of the quarter.
Who owns the agents when a second team arrives?
OwnershipIf the answer is "whoever wrote the prompt", the second team will copy it, change it and drift away from the first.
Try this on your organisation
Plan your first 90 days.
Fill in your first two teams and your first piece of work, and tick what's already agreed. The plan on the right updates as you type, with real dates. Nothing you type leaves this page.
- One team, one real piece of workShow: One item shipped, with the record of who asked, who approved and what it used.
- A second team, working its own wayShow: Two teams on one roadmap, and one agent improved from a retro.
- A view leadership can useShow:
This is a plan to run, not a promise. Some teams move faster and some slower; the order matters more than the dates.
With Wakalix
What each phase looks like in Wakalix.
Wakalix is built around the same steps, on the AI tools your teams already use. On the Pro plan, onboarding happens in your first week.
- 1Days 1–30: one team, on the recordSign-in uses your own identity provider. Nothing runs until someone on your side approves the plan, agents stop and ask when a decision belongs to a person, and a person accepts every piece of work.
- 2Days 31–60: a second team, its own wayTeams choose their harness, the agents they need and the rhythm they work at. Leadership still sees one roadmap, one set of guardrails and one record. Every retro feeds back into your agents.
- 3Days 61–90: a view leadership can useEvery item is estimated before it starts, and tracked against that estimate as it runs. The quarter's first answer comes from one record.

Your teams keep the AI tools they chose. Wakalix never runs the model.