Lines of Code got a better publicist

"75% of our code is now AI-written." Nearly every AI productivity headline this year quotes a volume number, and volume is the metric we spent decades learning to distrust. 

Percent-of-code-written-by-AI is just lines of code with a better publicist. These numbers influence budgets, performance expectations, and headcount decisions. I'll walk through why the popular claims can't fail (and therefore say nothing), what the outcome evidence actually shows, and a practical model for measuring AI-assisted engineering on capability rather than activity: what changed in the codebase, the team's workflows, and people's judgement.

Who is this talk for?

Engineering leaders and anyone whose team's AI use is being measured (or mis-measured): CTOs, EMs, staff engineers, and the platform/DevEx people asked to produce the numbers.

What will attendees take away?

  • How to spot unfalsifiable volume claims and adoption-ranked maturity ladders, and the question that deflates them: "is that an outcome, or a volume?"

  • What the research actually shows: the strongest pro-adoption evidence, the counter-evidence, and why self-report has stopped working as a measurement instrument.

  • A three-layer model for calibrating on observable capability (team adoption, individual mastery, craft signals) that you can apply to reviews and planning without inventing new vanity metrics.

David Curlewis headshot.

David Curlewis



He/Him
Senior Engineering Manager @ Canva

Dave leads AI Platform teams at Canva: evaluation, inference, data enrichment. Twenty-five years of engineering (including Trade Me, Timely, Sharesies, GitLab) taught him that most interesting problems are organisational, not technical.

These days he’s also interested in how AI impacts people, and how we can make the most of these technologies.

LinkedIn