Engineering rebuilt around AI agents

Context

A global consumer learning platform with millions of learners. Engineering was large and organised by function. Releases were slow, delivery was hard to predict, and adding people wasn’t adding speed.

What we did

• Reorganised engineering from functional teams into stream-aligned squads, each owning an outcome.
• Designed an agent-assisted delivery process: engineers and AI agents work from the same written intent, so agents can draft, test and review with the right context.
• Built the first agent workflows and context standards hands-on, then rolled them out team by team.
• Put delivery on DORA metrics with a weekly review, plus definition-of-ready and incident review practices.

How it ran

One squad first, then the rest. Leadership reviewed the numbers weekly, and the process changed when the numbers said so.

What changed

• Typical cycle time down to 3 working days, with a 0.3% change failure rate on daily deployments.
• Squads that needed 10 to 11 engineers now run on 5 to 6 and deliver more.
• Planned-versus-delivered predictability up from about 50% to 65%.

Client names stay private. The work is real, and references are available on request.