Coding agents consolidate: winners ship more, losers sunset

The winner's story: 20 times more, same team

CodeRabbit says it has shipped 20 times more software over the last 6 months without adding engineers, by using Codex, and that it uses OpenAI models to govern and trust AI-generated code. The company's chief executive, Harjot Gill, is quoted in the OpenAI material as saying the first bottleneck every company hits as soon as it adopts coding agents is code reviews. That framing is the story in one line: the agents can write the code, and the scarce resource is deciding what is safe to ship. m:…6535

CodeRabbit says it is used by over 20,000 organisations and 150,000-plus open-source projects, and that it helps teams understand, review and secure code produced by AI agents and decide what is ready to ship. Its latest offering, CodeRabbit Secure, uses Daybreak with Codex models to find and fix vulnerabilities, and the company says it found over 43,000 security vulnerabilities across 13,000 deep scans using Codex models. The company adopted OpenAI o1 and OpenAI o3 early to improve code review capabilities, and it now uses GPT-6 Luna for subagents and GPT-6 Astra for planning and orchestration, according to the same material. m:…6535

The loser's story: Cody sunsets

Sourcegraph announced the opposite move in the same week. The company says it will end of life Sourcegraph Cody on January 31, 2027, and it introduced Cody in early 2023, when AI-assisted coding looked very different. Sourcegraph says it decided not to evolve Cody into a coding agent two years ago, and that it knew the day would come when Cody no longer had a place in the market. That is an unusually frank admission: the company chose not to compete as an agent, and the market caught up with the choice. m:…3273

Sourcegraph is positioning itself as retrieval infrastructure for the agentic coding era, and it says it is building a deterministic code graph for entire codebases. The company says it is focusing on helping engineering teams make faster decisions, improving the quality and accuracy of AI-generated code, and executing codebase-wide changes and security fixes with confidence. The pivot is from owning the assistant to owning the map the assistants use: code search and the deterministic code graph help developers and agents find code, trace references and understand relationships across repositories, even in the largest codebases. m:…3273

Sourcegraph says its MCP, CLI and other APIs bring its code intelligence into the tools developers already use, including Claude Code, Cursor and other coding agents. Coding agents like Amp, Claude Code, Cursor and Codex can now take on complex tasks and write substantial amounts of code autonomously, according to Sourcegraph, and for teams using Cody for code understanding, Deep Search and MCP integrations can support many of those workflows through the coding agents they already use. Sourcegraph also says it will work directly with Cody customers to understand their requirements and plan transitions, though specific migration paths for enterprise customers with complex custom Cody-based workflows remain to be defined. m:…3273

The consolidation move: OpenAI buys Pierre

OpenAI announced an agreement to acquire Pierre Computer Company on November 4, 2026, and the deal has not yet closed and remains subject to regulatory approvals, with the price undisclosed. Pierre's open-source projects will remain open source following the acquisition, according to Qodo's account. The acquisition pulls Pierre's Trees and Diffs libraries into Codex, deepening the integration of the developer workflow inside OpenAI's own coding agent rather than leaving file browsing and diff editing to outside tools. m:…0847

The context is a rapid shift in who writes the first draft. More than 40% of pull requests on GitHub originated from an agent's first draft in September 2026, compared with 17% in March 2026, according to GitHub Octoverse 2026 as reported by Qodo. Yet coding agents currently rely on fragmented instruction files like AGENTS.md and CLAUDE.md, leading to rule drift, because each agent reads its own instruction file and no single place shows what every agent changed. Only 34.6% of surveyed developers report that coding agents always follow organisational standards, according to the Qodo State of AI Code Quality 2026 survey. m:…0847

The governance layer in practice

Packsize shows what that governance layer looks like inside a real engineering team. Packsize builds on-demand packaging systems that create right-sized corrugated boxes for customer orders, and its engineering environment spans roughly 50 microservices. AI coding tools increased the volume of pull requests at Packsize, shifting the bottleneck from writing code to reviewing it, and Nathan Hansen, the company's AI architect, says the AI used to write code is generally insufficient for evaluating its own implementation. That is the same point Qodo makes in the abstract, stated by the customer who felt it. m:…8557

Packsize mandates that every pull request undergoes a review by Qodo before production, and Qodo automatically extracts review standards from skill.md files in each repository to ensure compliance. Qodo uses multiple models to perform adversarial reviews of code, rather than relying on a single AI that might overlook its own mistakes, and Packsize says code review time now takes just minutes with Qodo. The company credits Qodo with preventing a production outage by flagging a compatibility issue in shared packages, saying the update would have caused a production outage without that safeguard. Packsize aims to move toward a software factory model with autonomous cloud-based agents, though the specific roadmap for that transition is unresolved. m:…8557