Source: https://internal-agents.com/agents/workos-project-horizon # WorkOS — Project Horizon An internal autonomous 'code factory' where a continuously running swarm of agents handles the implementation loop while engineers focus on requirements and acceptance testing. Deliberately modular so the harness can evolve. > This entry describes supporting infrastructure that other work builds on. The catalog classifies it as a platform. The record also reports a workflow. - Approach type: Platform - Deployment stage: Deployed - Autonomy: Drafts reviewed - Evidence strength: Detailed primary - Status: Internal - First reported year: 2026 - Work: Coding, Code review, Security - Interfaces: Linear, Github, Slack, Web - Invocation: Event driven, Interactive - Entry reviewed: 2026-08-31 ## Where people stay involved - **requirements and acceptance criteria → tested implementation** — Outcome review · Level 4 ## Overview ### Summary An internal autonomous 'code factory' where a continuously running swarm of agents handles the implementation loop while engineers focus on requirements and acceptance testing. Deliberately modular so the harness can evolve. Fact · Reported · High confidence · `workos-project-horizon--summary` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) - Supports · [2] [Applied AI Showcase (Horizon with Claude Remote Routines)](https://workos.com/blog/applied-ai-showcase) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-2/content.md) - Contextualizes · [3] [An autonomous UI-quality program](https://workos.com/blog/autonomous-ui-quality-program) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-3/content.md) ## How it works ### Supporting component Separate what runs code from what manages the lifecycle Fact · Reported · High confidence · `workos-project-horizon--primitives-0` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) ### Supporting component A runtime controlled end-to-end, with lifecycle APIs and egress controls for the threat model Fact · Reported · High confidence · `workos-project-horizon--primitives-1` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) ### Supporting component Each run ships work and produces the next set of fixes, surfacing where the platform is brittle Fact · Reported · High confidence · `workos-project-horizon--primitives-2` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) ### Supporting component Tuning tools is ongoing, not a one-time integration Fact · Reported · High confidence · `workos-project-horizon--primitives-3` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [2] [Applied AI Showcase (Horizon with Claude Remote Routines)](https://workos.com/blog/applied-ai-showcase) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-2/content.md) ## Supervision evidence ### Operating model assessment Level 4 for requirements and acceptance criteria → tested implementation; human attention boundary: outcome-review. Inference · Catalog judgment · High confidence · `workos-project-horizon--operating-models-0` Confidence reason: The source describes engineers focusing on requirements and acceptance testing while agents run the implementation loop. Qualifications: - Observation date: 2026-05-06 Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) ## Implementation details ### Sandbox Cloudflare Containers + Sandbox SDK; disposable, tightly scoped sandboxes with explicit lifecycle APIs and egress controls; full monorepo stack in Docker dev containers Fact · Reported · High confidence · `workos-project-horizon--architecture-sandbox` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) ### Harness Modular by design; the core article runs OpenCode in the sandbox; the Applied AI Showcase runs Claude Remote Routines. The harness is swappable as agent tech changes; separate PM, implementation, and prospective verification/security roles Fact · Reported · High confidence · `workos-project-horizon--architecture-harness` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) - Supports · [2] [Applied AI Showcase (Horizon with Claude Remote Routines)](https://workos.com/blog/applied-ai-showcase) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-2/content.md) ### Model Swappable; the harness is the constant, not the model Fact · Reported · High confidence · `workos-project-horizon--architecture-model` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) - Supports · [2] [Applied AI Showcase (Horizon with Claude Remote Routines)](https://workos.com/blog/applied-ai-showcase) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-2/content.md) ### Interfaces linear, github, slack, web Fact · Reported · High confidence · `workos-project-horizon--architecture-interfaces` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) ### Tool access A custom MCP server stitches internal data sources (Datadog, Sentry, Slack, WorkOS Pipes); all outbound traffic proxied through Workers with allowlists, limits, logging, and token injection Fact · Reported · High confidence · `workos-project-horizon--architecture-tool-access` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) - Supports · [2] [Applied AI Showcase (Horizon with Claude Remote Routines)](https://workos.com/blog/applied-ai-showcase) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-2/content.md) ### Knowledge AGENTS.md and CLAUDE.md capture scripts, docs, conventions; MCP codifies the patterns engineers already follow; Notion + Figma for specs/mockups Fact · Reported · High confidence · `workos-project-horizon--architecture-knowledge` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) ### Credentials WorkOS Pipes (no OAuth/token-refresh to maintain); scoped short-lived GitHub tokens per user; engineers use their own identity in the MCP; least-privilege + egress controls Fact · Reported · High confidence · `workos-project-horizon--architecture-credentials` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) ### Context management The orchestrator pauses/resumes sandboxes and tracks state + artifacts across a run Fact · Reported · High confidence · `workos-project-horizon--architecture-context-mgmt` Confidence reason: A linked first-party source states the claim. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) ## Lessons and interpretation ### Lesson You need purpose-built agent infrastructure; a runtime you control end-to-end with lifecycle APIs and egress controls Inference · Catalog judgment · Medium confidence · `workos-project-horizon--lessons-learned-0` Confidence reason: The catalog derives this observation from the linked sources. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) ### Lesson Separate concerns: sandboxes are an execution primitive; the orchestrator is the control plane Inference · Catalog judgment · Medium confidence · `workos-project-horizon--lessons-learned-1` Confidence reason: The catalog derives this observation from the linked sources. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) ### Lesson Build modularly so the harness can evolve; OpenCode today, Claude Remote Routines tomorrow, without rebuilding the platform Inference · Catalog judgment · Medium confidence · `workos-project-horizon--lessons-learned-2` Confidence reason: The catalog derives this observation from the linked sources. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) - Supports · [2] [Applied AI Showcase (Horizon with Claude Remote Routines)](https://workos.com/blog/applied-ai-showcase) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-2/content.md) ### Lesson Make autonomy a platform; the system gets faster and more reliable through use as fixes feed back in Inference · Catalog judgment · Medium confidence · `workos-project-horizon--lessons-learned-3` Confidence reason: The catalog derives this observation from the linked sources. Evidence: - Supports · [1] [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-1/content.md) ### Lesson MCP tuning is an iterative product, not a one-time integration; codify the patterns engineers already follow Inference · Catalog judgment · Medium confidence · `workos-project-horizon--lessons-learned-4` Confidence reason: The catalog derives this observation from the linked sources. Evidence: - Supports · [2] [Applied AI Showcase (Horizon with Claude Remote Routines)](https://workos.com/blog/applied-ai-showcase) · [Preserved copy](https://github.com/steel-experiments/internal-agents-map/blob/main/archive/sources/workos-project-horizon-source-2/content.md) ## Sources 1. [Project Horizon - an autonomous code factory at WorkOS](https://workos.com/blog/project-horizon) - Engineering blog · First party · Evidence - Original URL: - Accessed: 2026-08-12 · Last verified: 2026-08-31 - Preserved copy in the repository: 2. [Applied AI Showcase (Horizon with Claude Remote Routines)](https://workos.com/blog/applied-ai-showcase) - Engineering blog · First party · Evidence - Original URL: - Accessed: 2026-08-12 · Last verified: 2026-08-31 - Preserved copy in the repository: 3. [An autonomous UI-quality program](https://workos.com/blog/autonomous-ui-quality-program) - Engineering blog · First party · Evidence - Original URL: - Accessed: 2026-08-12 · Last verified: 2026-08-31 - Preserved copy in the repository: 4. [Hacker News submission for Project Horizon](https://news.ycombinator.com/item?id=48039227) - Hn thread · Community · Discovery - Original URL: - Published: 2026-05-06 · Accessed: 2026-08-12 · Last verified: 2026-08-31 - Preserved copy in the repository: