Jay Hack presents Codegen 3.0, a platform for AI coding agents that can search codebases, create pull requests, and run automated fixes. It integrates with GitHub, Slack, Jira, and other tools, with features like custom API keys, multiple model choices, and transparent analytics for enterprises.
Chapters
CodeGen essentially applies AI to the field of software engineering. We're going after very large codebases, getting AI to make significant changes. The vast majority of code that is written in the future, especially for large organizations, will be done by AI. CodeGen is a platform for running code agents at scale that can search code, edit code, create pull requests, and interact with you across Slack, Jira, Linear, GitHub, and more. CodeGen responds to comments and reviews on its pull requests just like humans do, adding additional commits to its pull requests. In response to user feedback. When tests fail, CodeGen spins up an additional agent to submit a commit that will fix the broken check so it can iterate to the goal line without human intervention. It integrates tightly with dozens of first-party integrations, including Figma, Linear, ClickUp, Postgres, and more, so you can have it perform live analytics queries against your production database, create detailed Notion documentation based upon your codebase, or have it burn down your backlog directly from Linear. CodeGen is built for scale and with enterprise configurability in mind, including custom API keys, dozens of model choices, arbitrary MCP servers, and fine-grained codebase dependency management. It even supports AI code review on all human pull requests so you can spot issues before they get to production and maintain code quality across your entire team. Most importantly, CodeGen is an open platform providing transparent analytics on cost, agent impact over time, as well as a detailed API to retrieve all agent traces and associated data. We think of ourselves as an operating system for code agents, so just like a CPU has many things that surround it to make it effective. CodeGen is like that for code agents.
