Dark factories: why your AI coding setup is already outdated

In this episode, I sit down with Chris Weichel, CTO and co-founder of Ona (formerly Gitpod), to explore how the company evolved from cloud development environments to building infrastructure for background agents — autonomous AI processes that pick up work, write code, open PRs, and even auto-approve them based on risk assessment. Chris walks us through the three stages of AI adoption in software engineering: better autocomplete → the software conductor (human orchestrating agents) → background agents (the factory). We dive deep into how Ona handles governance and automated code review with a risk ladder approach, and how their team of 20 engineers dealt with 450 open PRs when agents started shipping code. We also discuss the future of IDEs (spoiler: terminal agents' days are numbered), the cognitive load of parallel development, and why software engineering remains a deeply human endeavor even as agents accelerate every phase of the SDLC. 🔗 Resources & Links: • Ona (formerly Gitpod): https://ona.com/ • Background Agents: https://background-agents.com/ • "The Rise of the Software Conductor" (Ona blog): https://ona.com/stories/time-between-disengagements-the-rise-of-the-software-conductor • "The Software Conductor's Handbook" (Ona blog): https://www.gitpod.io/blog/software-conductors-handbook • Chris Weichel on GitHub: https://github.com/csweichel • Book recommendation — "The Origins of Efficiency" by Brian Potter (Stripe Press): https://press.stripe.com/origins-of-efficiency • Peter Steinberger / OpenClaw (guest recommendation): https://github.com/steipete 📚 Also mentioned: • The Phoenix Project & The Unicorn Project (Gene Kim) • LM Studio & Ollama for local models • MLX on Apple Silicon ⏱️ Chapters: 00:00 — Introduction & Chris's background 01:40 — From Cloud Development Environments to Background Agents 05:00 — The three stages: autocomplete → conductor → factory 08:15 — Running agents in isolated environments for security 11:00 — The PR explosion: 450 open PRs with 20 engineers 15:30 — Automated code review as a risk ladder 19:00 — Governance: earned trust, staged rollout, and SOC 2 compliance 22:00 — Multi-agent review with meta-reflection 25:00 — Non-determinism: agents vs. humans at upholding standards 28:30 — Background agents in production: CVE mitigations, observability loops 32:00 — Keeping architecture under control in the age of agents 34:00 — The human element: "We care about humans writing software" 36:00 — Cognitive load and the "five slot machines" problem 40:00 — The future of IDEs and terminal agents 42:30 — Closing: background agents are a reality for regulated enterprises 44:00 — Book recommendation & what's next for Chris 🎙️ AWS Developers Podcast Subscribe for conversations with the builders shaping the future of software development.