Tom Adams’ Post

Two months is nice, zero incidents on payment paths is the real result. Curious what the safety net looked like – that's the part other teams can't copy from a blog post.

The part about the bottleneck moving from code generation to code review is probably the most interesting one for me. We see the same thing in delivery all the time: make one part of the system faster and the constraint simply moves somewhere else. So the 48% increase in merge requests is impressive, but understanding where the next bottleneck appears may be even more valuable. That's where productivity starts becoming a system question, not an individual one.

Tom, this connects AI-assisted coding with systems engineering: higher developer throughput only matters when architecture, testing, deployment, and operational controls scale alongside it. Velocity and reliability have to move together

Very strong result. Doing this kind of refactor in a live payments environment with zero incidents says a lot about the review and ownership around it. In enterprise codebases like this, boring correctness is what actually builds trust in AI assisted delivery.

Really nice write up, and sounds like some very solid results, too. On the risk tagging of MRs point, we've been thinking about similar, specifically overlaying the production context we have from everything happening in incidents. I'm pretty bullish on the idea that we'd be able to flag risks in code changes that look syntactically correct and pass a human review, but fail to factor in the production environment where they're being deployed, and the historical context of changes in that area. Would love to trade notes sometime!

TIRED Skeletons we need to fix Token budgets WIRED Modernisation opportunities No blush policy

From personal experience working with Adyen engineers, I can attest to the very high caliber of talent :)

Very interesting! Thank you Tom Adams 🙏

💪🔥