Two engineering leaders explain what a year of deploying deterministic AI tooling at enterprise scale actually taught them, including why AI-generated code is creating a new code review bottleneck industry-wide.
Dov Katz (Developer Experience) and Khalid Elsawaf (Domain Architect, Secured Financing Technology) share a one-year progress update from Morgan Stanley: 5,000 users onboarded, 2,000+ running recipes, 1,000+ commits shipped, and 250 community-contributed recipes in the company’s internal marketplace. They cover why adoption at scale takes sustained effort rather than a one-time launch, why people adopt experiences rather than tools, and why a fix isn’t done until it’s merged: generating pull requests at AI speed just moves the bottleneck to code review, and what gets changes into production is confidence backed by evidence, not the number of PRs opened. They also walk through matching review to risk, with the data-driven risk assessment model first built for DevOps change management now shifting left into the PR process itself and FINOS work underway to bring it to the wider industry.
They close with two live examples: an AI-assisted AspectJ vulnerability remediation that pulls changelogs and builds a custom OpenRewrite recipe on the fly, and a fully AI-orchestrated Java build cache rollout across thousands of repositories, including automated PR triage, CI log analysis, and confidence scoring before a human ever sees the change.
