|
April 23, 2026
Stripe, Harvey & more on background agents
This issue focused on the first Background Agents Summit, the latest milestones from software-factory.dev, and a new Veto update for executable discovery across container layers.
The throughline was practical adoption: how teams move from local coding assistants to background agents that can safely run in shared infrastructure, ship work continuously, and still fit into human review loops.
TL;DR
|
|
Background Agents Virtual Summit, May 6: 11+ speakers from teams running agents in production.
Last call for talks: the summit CFP was closing that Sunday.
software-factory.dev: two weeks in, the finale was streaming that Friday.
Veto: executable discovery now worked automatically across container layers.
|
Background Agents Summit
The summit was positioned as a practical event for engineering teams already asking the hard operational questions: where agents should run, how to review their work, what use cases to tackle first, and how to avoid adding overhead instead of removing it.
Speaker highlights from the issue included Stripe, Harvey, Monzo, PostHog, incident.io, WorkOS, AWS, OpenClaw, Falco, and Ona's own software-factory.dev work. The agenda emphasized context engineering, sandboxing, security, identity, review workflows, and proving value in production.
Event
A production-first agent lineup
Sessions called out in the issue included Stripe on 1,000+ agent PRs a week, Harvey on moving agents off laptops and onto shared cloud runtimes, Monzo on running agents inside a bank, and Ona on building software-factory.dev in public.
View the recordings.
|
software-factory.dev, Two Weeks In
The newsletter framed software-factory.dev as a live experiment in agentic engineering: one empty repo, a product spec, a set of automations, and a strict rule against human-written code.
At that point the factory had merged 302 PRs, written 49,002 lines of code, and was running 14 automations from planning through deployment. The issue also called out a less obvious lesson: once agents can ship code continuously, the human role shifts from writing code to designing the system around quality gates, context, and feedback loops.
The hardest week-two lesson was taste. The system was fast and reliable, but product judgment still depended on humans. That became a major theme for the finale stream and the week-one recap.
Read the week-one recap or watch the finale.
Veto Update
The Veto update in this issue was focused on reducing manual configuration: executables could now be discovered across container layers in real time and matched by content hash, so users no longer had to enumerate them by hand. Read the update.
Events
Software Factory finale: the closing episode of the two-week experiment, live that Friday.
Background Agents Virtual Summit: May 6, with registration and CFP links front and center.
Reuters Momentum AI, New York: April 28-29, with Ona's CTO joining an enterprise agentic AI panel.
Wishing you and your agents a mercifully dull week,
Lou
Field CTO, Ona
|