Helping an engineering team work with AI agents

The challenge

AI coding tools were getting better every month, but most of the team used them like a smarter autocomplete. Tickets that needed one or two days of real work still took two to four weeks to get through the process, because of hand-offs between product, development and QA. Faster typing wasn’t going to fix that. The way the work was organised had to change.

What we changed

• Builders instead of FE/BE/QA silos. Each “Product Builder” owns a work package from the problem through to release, with AI agents doing most of the implementation.
• Guardrails as code. Project context files, coding rules and review checklists live in the repositories, so every agent session starts with the team’s standards already loaded.
• Non-engineers in the codebase. Product managers and designers prototype and change things directly, instead of writing specs and waiting.
• Coaching, one person at a time. The most experienced developers resisted the change most. They were won over by pairing on real tasks, not by mandates.

Results

The team now rarely writes code by hand. Engineers spend their time on specifying, reviewing and verifying, which are the parts that actually decide quality. I described the journey publicly in an interview in the “Interviews with technology leaders” series on the YouTube channel AI with Michal Juhas.

What I’d tell another CTO

Start with yourself. You can’t lead an AI transformation you haven’t lived through. Use the tools daily on real work first, then bring the team along with concrete wins rather than slide decks.