
Date & time
17:00
Register for the panel discussion
Login or join LeadDev.com to view this content
Every manual approval gate in your pipeline is a bet that a human will make a better decision than your automation. At Booking.com’s scale, that bet stopped paying off. Engineers were breaking out of focused work to glance at a dashboard and click approve, over and over, every day. Verification quietly became vibes.
So before going anywhere near agents, Varun Purohit’s platform team measured the problem. They defined a metric DORA doesn’t give you: MARD, the wall-clock time a deployment spends waiting for a human to decide.
Then they went after it with continuous verification: automated checks that run on every deployment by default. They proved it out on a deliberately mixed set of services, spanning criticality tiers and business units rather than the easy cases, and the time deployments spent waiting on a human fell away. Many of Booking.com’s services now deploy with no manual approval at all.
Varun, who leads build and deploy at Booking.com, will walk through how they got there: the checks that turned out to be enough, the ones that weren’t, what broke during rollout, and why none of this is a detour on the way to agentic deployment. AI is a multiplier. It magnifies clean signals and tribal knowledge alike, which is why the order of the work matters.
Key takeaways:
- How to measure what manual approval gates cost you, and why your DORA metrics won’t show it
- Which approvals you can replace with automated verification, and which signals have to exist before you do
- Why platform-defined health checks stop scaling, and how handing signal definition to service teams unlocked adoption
- How to place your organization on a staged path from manual approval to autonomous agents, and why trust breaks when you skip a step