--- title: "Factory's Missions Let Droids Build Code Autonomously for Two Weeks" subtitle: "Factory's Missions feature lands on Desktop, letting autonomous Droids run multi-day engineering tasks while your laptop is closed" url: "https://alphasignal.ai/news/factory-s-missions-let-droids-build-code-autonomously-for-two-weeks" author: "AlphaSignal Newsroom" author_url: "https://alphasignal.ai/editorial-team" company: "Factory" company_url: "https://alphasignal.ai/company/factoryai" published: "2026-06-10T17:41:12.000Z" updated: "2026-06-10T18:41:36.449Z" type: "News" topics: - "Development" - "Infra" subtopics: - "Code Generation" - "Model Serving" - "Inference Optimization" upvotes: 617 read_time: "5 min" source_url: "https://factory.ai/case-studies/chainguard" comments: 0 --- # Factory's Missions Let Droids Build Code Autonomously for Two Weeks > Factory's Missions feature lands on Desktop, letting autonomous Droids run multi-day engineering tasks while your laptop is closed Most AI coding tools are built around a conversation. You type, the model responds, you review, repeat. [Factory's Missions](https://factory.ai/product/missions) are built around something fundamentally different: you describe a goal, close your laptop, and come back to a finished pull request. That workflow just got a lot more accessible. Missions are now available on the Factory Desktop app, bringing multi-day autonomous engineering tasks to the same interface where you already run your Droids. ## The problem every long-running agent hits Single sessions hit limits. Context windows fill up. Attention degrades over long trajectories. Droid starts forgetting what it already tried, re-reading files, losing track of the bigger picture. The naive fix is to just run multiple agents in parallel, but coordination is hard. Agents conflict, duplicate work, and drift without structure. Agent sessions work well for focused tasks, but most real projects are too broad and complex for a single context window to hold. A single agent eventually runs into a problem: the more it sees, the less focused and reliable it becomes. Factory's answer to this is a three-role architecture that sidesteps the problem entirely. ## How the machine actually works An orchestrator plans and decomposes the user's goal, steering execution to completion and passing all validation gates. It avoids accumulating overly granular context, delegating all investigation and implementation to subagents and workers. It doesn't drive validation directly -- the system injects validators at milestones to surface gaps. Think of it as a three-tier hierarchy: - **Orchestrator:** Plans the project, breaks it into milestones, and manages the overall trajectory without getting bogged down in implementation details. - **Workers:** Complete well-specified features with clear success criteria. They iterate until they believe the work is correct, then hand it off. But the final judgment on correctness is not their call. - **Validators:** Every milestone ends with a validation phase: workers review the accumulated work, run tests, check for regressions, and verify that everything integrates. When validation surfaces issues, the orchestrator creates follow-up work to fix them before moving on. The key insight is that agents are highly reactive to their context, and an agent's trajectory is append-only, so the model's reasoning at any given point is a function of every past thought, observation, and action. By giving each sub-agent a narrow, fresh context, Missions avoids the degradation that plagues long single-agent runs. ## The compaction engine underneath it all The other half of what makes multi-week sessions possible is Factory's context compaction engine -- the system that decides what to remember and what to compress as a session grows. Factory developed an evaluation framework to assess context compression strategies for AI agents working on extended software development tasks that generate millions of tokens across hundreds of messages. Testing on over 36,000 production messages from debugging, code review, and feature implementation sessions, Factory's structured summarization approach scored 3.70 overall compared to 3.44 for Anthropic and 3.35 for OpenAI, demonstrating superior retention of technical details like file paths and error messages. The reason this matters in practice: coding sessions have state. The agent needs to know not just facts from the past, but the current state of the task: what has been tried, what failed, what is left to do. Generic summarization often captures "what happened" while losing "where we are." Factory's approach merges new summaries into a persistent state rather than regenerating from scratch, so critical details don't drift across compression cycles. ## Droid Computers: the infrastructure layer Missions on Desktop are paired with Droid Computers -- Factory-managed remote machines that keep running after you disconnect. Each Droid gets a persistent compute environment that survives between sessions. Installed packages, cloned repos, credentials, and running services carry forward. You can delegate a Mission to a Droid Computer, close your laptop, and the agent keeps working autonomously. Factory offers two options: Cloud Computers, which are managed machines provisioned through the app with SSH access and instant resume, and BYO Machine, which turns existing hardware into a Droid Computer. A Droid can open a staging environment in a browser, click through a user flow, report failures, switch to VS Code, run extension commands, read output, and act on it. ## A real-world stress test: two weeks, 80 packages The Chainguard case study is the most concrete evidence of what Missions actually unlocks at scale. [Josh Wolf, Staff Engineer at Chainguard](https://factory.ai/case-studies/chainguard), ran a single Factory session for two weeks across six repositories and shipped 80 packages. His team works across large codebases with a constant influx of upstream package updates and a rigorous engineering review process that demands context from each step. His take on the compaction difference: "When you don't have to think about context windows, you can treat Droid like a colleague that just remembers what you've been talking about." He had tried every memory solution the ecosystem offered -- memory MCPs, custom SDKs, the rest -- and found none of them delivered. The thing that changed his mind was that Factory's compaction "just works. ## Sources - [X post by @factoryai](https://x.com/factoryai/status/2064764834928107914) - [Factory Missions | Multi-Agent Orchestration](https://factory.ai/product/missions) - [Two Weeks, One Session, Endless Packages | Factory.ai](https://factory.ai/case-studies/chainguard) - [Introducing Missions | Factory.ai](https://factory.ai/news/missions) - [How Missions Work | Factory.ai](https://factory.ai/news/missions-architecture) - [Factory Pricing](https://factory.ai/pricing)