the self-driving codebase series just wrapped. start here.

Over the past four weeks, we ran a series on what it actually takes to move from coding agents on your laptop to background agents running autonomously in production. The infrastructure, security, and use cases where agents deliver value today.

CTO Chris Weichel and I opened the series by defining what separates a background agent from a regular coding agent: it picks up work by itself and operates autonomously until a human needs to step in. To make that real, you need three infrastructure primitives: compute, events, and governance.

Lucas from our solutions engineering team walks through the experiences of Fortune 500 teams in a live CI migration demonstration: one engineer crafting the prompts and scripts, then fanning that work out across hundreds of repositories with a fleet of AI software engineers. Each one picks up a repo, runs the migration, validates it, and submits the result for review. With automations, headcount stops being the bottleneck.

Over 40,000 CVEs were published last year, and the current lag from discovery to database entry is around three months. Agents are well-suited to this kind of high-volume, well-scoped remediation: pick up a vulnerability, patch the affected repos, run the tests, submit for review.

The final session brought in Sunil Divvela, a mainframe modernization specialist from AWS, alongside CTO Chris Weichel to tackle legacy code nobody documented. They showed how fleets of AI software engineers can explore COBOL, extract structured specs, and verify the migrated code behaves the same way. That verification step is where most modernization projects stall.

Four sessions covered the primitives, the use cases, and the security model. But there's one more piece we've been working on. Something that maps the entire space. Look for it on background-agents.com next week.

May your PRs merge continuously,

Lou
Field CTO, Ona