Put your coding agents in the outer loop
The inner loop is your IDE, the outer loop is CI, review automation and scheduled upkeep. Most teams put agents in the inner loop, but the work that compounds lives one loop out.
Written by
-
Benedikt Stemmildt Founder & Co-CEO
Published: 23 July 2026 Updated: 2 min read
Most teams have their coding agents running in the wrong loop.
Inner loop and outer loop
The split is older than agents. The inner loop is your IDE, your local environment, your immediate feedback cycle. The outer loop is CI, code review automation, issue triage, dependency updates, anything that runs in the cloud on a schedule.
Most engineers I see reach for the inner loop first. It feels familiar. Local code, local files, local credentials, an answer the moment you hit enter.
The work that compounds runs on a schedule
But a lot of the work that actually compounds lives in the outer loop. Dependency bumps, lint cleanups, test coverage walks, stale issue triage, doc upkeep. None of it is glamorous. All of it is the work no one schedules, and it repeats forever.
The outer loop is also the safer place
The outer loop is also the safer place to run an agent. You can scope it to exactly the credentials the job needs, sandbox where it runs, and keep a full audit trail of what it did and why. An inner-loop agent inherits whatever your laptop has: your shell, your tokens, your blast radius if something goes sideways.
What it looks like in practice
If you want to see this in practice, the Gitpod team keeps their outer-loop automations in the open on GitHub (opens in a new tab), one inspectable file each.
A PR reviewer, a post-merge verifier, a stale-issue reviewer, a bug fixer, each a small config you can read end to end. Not a fleet you have to take on faith. A directory you can open, audit, and copy the pattern from.
Start with one boring job
That is the shift worth teaching your team. Not “can an agent help me write this function”, but “which recurring, boring, well-defined jobs can I hand to a scoped agent and check on later”. Start with one. Give it the narrowest credentials that work, a sandbox, and logs you can read. Then add the next.
Most teams are still polishing the inner loop. The bigger gains are one loop out.
Further reading
Written by
-
Benedikt Stemmildt · Founder & Co-CEO
More on Software Factory
- Six developer experience investments that outlast the next model Most agentic tooling bought this year will not survive the next model generation. Six fundamentals will, because what is good for human developers turns out to be good for agents.
- What is a Software Factory? A Software Factory coordinates agents, context, deterministic gates, and human decisions across the path from intent to production.
- Build an auditor that improves your automations' prompts An automation that rewrites other automations' instructions sounds risky. Built as a loop of real runs, explained verdicts and a reviewed diff, it is continuous improvement with one gate.
Scaling needs named owners
Coding agents scale across teams when named people own the platform, the team practice, the guardrails, and the measurement. Rollout & scale is where that ownership is built and handed to internal owners, in your stack.
Training on this topic: Agentic Software Factory: Advanced Team TrainingWant to check one team first? The self-check: twelve yes or no questions, three minutes, instant result, no account. Take the self-check