{"slug": "how-we-accidentally-built-a-self-improving-software-factory", "title": "How we accidentally built a self-improving software factory", "summary": "Mutagent Cloud reports that its software factory Architech, a multi-agent system built on Pi, now runs in Mutagent Cloud's agent sandboxes and is being optimized by Helix, the agentic coding harness Architech itself built. Helix moved from laptops to a cloud sandbox on Mutagent Cloud so it can continuously monitor agent traces and actions, then was pointed back at Architech to optimize the architect, orchestrator and developer agents. The company says the closed loop compounds — a better Architech builds a better Helix, which makes Architech better again — but that the open problem is containing amplification, since a bad mutation to the developer agent makes every later iteration worse.", "body_md": "# How we accidentally built a self-improving software factory\n\nOur software factory, Architech, built our agent factory, Helix. Then we pointed Helix at Architech. Now the factory that builds our product is being improved by the product it built.\n\n## We built a software factory to build our product\n\nWe put our coding infrastructure together so that it builds software automatically, the way a software factory does. Internally we call it Architech. We use Architech to build our product for agent building and optimization, Helix.\n\n**Architech** is our software factory. It is a multi-agent system with parallel workers, built on Pi, and it takes a feature through five phases:\n\n- **Plan.** The architect turns requirements into a design and tasks, then publishes the tasks as GitHub issues.\n- **Build.** The orchestrator dispatches issues to agents in parallel, or plans and acts in a single pass for small changes.\n- **Review.** The developer reviews the issue or the pull request.\n- **Ship.** The orchestrator runs review, CI and merge, then watches the rollout.\n- **Monitor.** The orchestrator watches CI/CD, deploy health and the product.\n\n**Helix** is our agent factory, an agentic coding harness optimized for building agents. It is also built on Pi. It takes an agent through six stages: Spec, Build, Evaluate, Diagnose, Optimize, Ship.\n\nArchitech built Helix for months. That was the whole relationship: one factory produced the other.\n\n## Helix moved to the cloud and learned to watch agents\n\nHelix started out running on our laptops. We moved it into a cloud sandbox on Mutagent Cloud, so it can continuously monitor agent traces and agent actions.\n\nThen we pointed it at Architech. As Architech is a multi-agent system with parallel workers, Helix performs its job of optimising and watching its agents.\n\n## We put the optimization loop on our own software factory\n\nOnce Helix was watching Architech, we diagnosed it and put the optimization loop on it. When we program with Architech now, Helix optimizes the agents doing the work: the architect writing the design, the orchestrator splitting issues, the developer reviewing the pull request.\n\n## Architech moved to the cloud too\n\nMoving Architech to the cloud was the easy part, because Architech had already built the infrastructure for Helix. Now the same infrastructure runs Architech, the factory that built it. The setup is the same as on our laptops, only the coding harness is different. Both factories now run in Mutagent Cloud’s agent sandboxes.\n\n## The loop closes\n\nThis is the part we did not plan.\n\nEach turn of the loop compounds. A better Architech builds a better Mutagent Cloud and a better Helix. A better Helix makes Architech better again.\n\nWhat we ended up with is a self-improving software factory that is also very good at building agents. Agent building became a subset of the software factory, because the agent factory now lives inside it. The combined loop behaves like an engineer and an AI engineer working side by side.\n\n## Your coding setup is a multi-agent system too\n\nIf you run a coding agent with a spec step, parallel workers and a review step, that setup is a multi-agent system. It can be evaluated, diagnosed and optimized like any other agent.\n\n## The open problem is amplification\n\nA system that improves the thing that builds it can also amplify its own mistakes. A bad mutation to the developer agent makes every later iteration worse, including the review of the next mutation.\n\nWhat we still have to solve is how to contain that amplification: the guardrails that keep the system in check.", "url": "https://wpnews.pro/news/how-we-accidentally-built-a-self-improving-software-factory", "canonical_source": "https://mutagent.io/blog/self-improving-software-factory/", "published_at": "2026-09-29 18:50:53.589368+00:00", "updated_at": "2026-09-29 18:50:55.501601+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools", "mlops"], "entities": ["Mutagent Cloud", "Architech", "Helix", "Pi", "GitHub"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/how-we-accidentally-built-a-self-improving-software-factory", "markdown": "https://wpnews.pro/news/how-we-accidentally-built-a-self-improving-software-factory.md", "text": "https://wpnews.pro/news/how-we-accidentally-built-a-self-improving-software-factory.txt", "jsonld": "https://wpnews.pro/news/how-we-accidentally-built-a-self-improving-software-factory.jsonld"}}