Ryan Cooke argues that counting AI-generated pull requests can conceal whether a software factory helps a team deliver useful features. WorkOS's initial sandbox and coding-agent setup did not produce a clear improvement over engineers using coding agents locally, so the team began encoding more of its product engineering process into the system.
WorkOS separates TARS, which connects users and project activity across Slack, Linear, and GitHub, from Horizon, which orchestrates coding infrastructureAgent orchestration coordinates AI agents, tools, people, tasks, state, and control flow so a larger workflow reaches a verified outcome.. Dependency-aware tickets, a reviewed project brief called a hilltop document, and a planning agent let the system draft work, respond to approvals, and pick up later tasksWorkflow automation uses software or AI to complete a connected sequence of routine steps with less manual effort. while engineers remain able to refine scope and intervene.
Ryan Cooke describes an internal context engineAn AI context engine gathers, reconciles and ranks relevant information so an AI agent can act with the history, decisions and permissions surrounding a task. that uses an MCPModel Context Protocol is a standard way for AI applications to connect to external tools and sources of context. gateway to guide agents through company tools and data. He presents bug triage, support requests, durable organizational memoryAI agent memory is stored information that an agent can retrieve and use across steps, sessions, or changing contexts., and factory self-improvement as uses or next steps, while proposing customer delivery, defect rates, recovery time, and engineer adoption as more meaningful measures.
Watch on YouTube
