TechLead Conference

TechLead Conf London 2026: Adopting AI in Orgs Edition

Every team measures the engineering part of delivery - and it's already the fastest part. We measured the whole thing, from idea to learning, using data we already had: issue changelogs, structured experiment records, engineering analytics. We reconstructed the real clocks of our delivery and learned that an idea takes almost six months to become a validated result. Engineering we could see in detail. Much of the rest we couldn't see at all: some steps leave no trace anywhere, and between the steps there are hand-offs - unstructured, owned by nobody, impossible to measure. That's where the time lives. I'll walk through the investigation, the arithmetic behind why speeding up engineering alone barely moves the needle, and what actually does: improving the whole loop. Then the model we built on that premise and are piloting now - every step has a typed input and a typed output, and inside every step there's a self-checking improvement loop: AI drafts, a human who didn't draft it judges, and the judges themselves are reviewed - including the steps owned by analysts and medical reviewers, not just engineers.

Andrei Malyhin