The problem behind the task

It's surprisingly easy to get absorbed in doing a task well and forget to ask what the work is actually for. A few recent projects have reminded me to step back.

From an SBOM to a fleet-level question

I started with security work around software bills of materials: understanding what goes into a build.

But stepping back raised a more useful question: when a vulnerability appears, how do we connect that inventory to what's actually running across a cloud fleet?

The gap wasn't just about producing another document. It was about connecting build-time information with live software state so it could help with vulnerability management. That broader question shaped the Runtime SBOMs talk that was selected for the OCP Global Summit.

From a review agent to the workflow around it

I started building a small code-review agent. Then I began looking at everything that happens around a review.

A finding is useful, but what happens next? Which tests should run? Who interprets the results? How does that evidence feed back into the next change?

That led me toward an agentic test harness: specialized agents working together to analyze changes, select or generate tests, run validation, and interpret results. The interesting problem became less about one clever agent and more about the handoffs between them.

From a small inconvenience to a shared tool

With lab reservations, I noticed a small gap in how engineers found and coordinated shared resources.

Looking beyond the immediate inconvenience made the broader problem clearer: the coordination itself needed a simpler, self-service workflow.

I vibe-coded a first version of a reservation app in about an hour. The speed was satisfying, but the more useful part was finding something small enough to build that addressed a problem beyond my own immediate task.

I don't think the lesson is to turn every small task into a platform. Sometimes the small fix is exactly right.

It's to pause long enough to ask: am I improving this step, or is there a better way to solve the problem around it?

Watch the OCP talk or read my OCP reflection. Related project: Lab Reservation.

Written with AI assistance, grounded in my own work and reflections.