Selected experience
Identity verification platform: agentic AI strategy and engineering operations
North American identity verification SaaS unicorn, ~300 engineers
Agentic AI direction was needed on the roadmap while engineering operations had to become a permanent function rather than a temporary fix.
Outcomes
- Two agentic AI products shipped into a ~300-engineer organisation
- Engineering velocity raised ~50%; delivery-commitment reliability improved ~15%
- Engineering operations established as a permanent function
- 12 teams aligned on a shared milestone cadence within 8 weeks
Engineering operations became a permanent function, not a temporary fix.
Locate → Land → Live, retroactively, is the shape of the work that was done.
A North American identity verification SaaS unicorn with roughly 300 engineers needed agentic AI direction that could survive the jump from exploration to delivery — and an engineering operations function that would outlast the person building it. The founder served as Director of Engineering Operations at the time, not as A&O.
Locate meant finding where AI would change the business rather than the toolset. In an engineering organisation of that size, the expensive problems are rarely in the code. They are in the gap between what teams commit to and what ships, and in a quality feedback loop that tells you too late.
Land produced two agentic AI products shipped into the live organisation:
- A Product Agent that ran live discovery sessions with product managers, generated PRDs, triaged epics, and produced architecture decision records
- A QA Agent that cut post-release defect and hotfix rates
Alongside those, an AI QA assistant handled risk identification, test-plan authoring, and automated test-case generation — expanding effective QA capacity across every development team without expanding headcount. The QA function itself moved from a manual discipline to a quality-engineering one, with shift-left practices and risk-based quality gates built into CI/CD.
None of that would have held without the reporting layer. A CTO-level delivery analytics dashboard, built on automated Jira pipelines, converted raw delivery data into signal on throughput, commitment-versus-completion, and delivery health — replacing hero-driven status reporting with something systemic.
Live meant the operating discipline surviving real delivery pressure. Engineering velocity rose roughly 50% and delivery-commitment reliability improved about 15%, through dependency management, process redesign, and tooling automation rather than through pushing people harder. Release management became a core function, including SOC 2-compliant release processes.
Twelve teams aligned on a shared milestone cadence within eight weeks. Engineering operations was established as a permanent function. Agentic AI direction landed on the roadmap within twelve months.
The pattern behind all of it is the one we keep returning to: pilots stall when nobody owns the day-two work. Both agents shipped because ownership, sequencing, and measurement were designed in from the start.