The team
Four people: an architect, a developer, a product manager and a platform engineer. They build a multi-tenant SaaS product that was already in production, so every change had to fit an existing codebase and existing decisions.
Where they started
One large ask had been on the roadmap for almost 18 months. Everyone agreed it mattered, and nobody had the time to pick it up.
What they did
- Brought in what they already had. Their roadmap and the decisions already taken came in at the start, so the agents didn't have to rediscover them.
- Kept their own process. Sprints, roles and reviews stayed as they were. The agents worked inside them.
- Worked on one board. People and agents took items from the same board. When an agent hit a question only a person could answer, it stopped and asked.
- Accepted every change. Nothing an agent wrote landed until a person said yes.
The result
| Measure | Result |
|---|---|
| Work items finished, first three weeks | About 170 |
| Work items finished, first four weeks | 264 |
| Team | Four people |
| Waiting on the roadmap before | Almost 18 months |
| Finished in a single attempt | 73% |
| Finished within their token estimate | 94% |
| Share of what the agents read that came from cache | 95% |
| AI plan | A basic $25 Claude Enterprise seat each |
| Highest use of a person's plan limit | 63% |
| Times anyone hit their plan's limit | None |
| Changes accepted by a person | Every one |
All of it ran on a basic $25 Claude Enterprise plan per person, with no one ever hitting their limit: the busiest person peaked at 63% of theirs. Figures from the team's own Wakalix records, 21 October to 18 November 2026; plan-limit figures from the weeks Claude Code reported them (from 9 November).
In their words
"It's our process. We know how to work with it. There was no new learning curve at all because of the coding agents. It's like we onboarded coding agents onto our process."
The product manager expects the gains to keep growing as the team settles in. The CEO's favourite part was the leadership view: seeing delivery as it happens, without asking for a report.
What's next
The company plans to bring its other products onto Wakalix.
Why it worked
The agents started each item with the context it needed, questions went to the person who could answer them, and the team reviewed work as it arrived instead of in a pile at the end. That is the pattern behind the 5-4-3 promise: faster delivery, fewer tokens, less rework. See the 5-4-3 promise and the walkthrough Start to finish.