September 2025 Summaries

Zed is transitioning its AI pricing model from prompt-based to token-based to better align with the actual costs of running AI services, introducing models like GPT-5 and Gemini 2.5 to its offerings. This change aims to simplify pricing, reduce costs, and allow users to pay based on specific usage, adding a 10% markup to cover infrastructure and support expenses. To accommodate this shift, Zed offers a 14-day Pro trial with token credits and a three-month migration period for existing Pro customers. The new structure allows more flexibility, including options to use third-party APIs or local models, while maintaining a focus on developing Zed as a leading code editor. Existing users will be gradually transitioned to the new model, and alternatives for AI integration are provided, ensuring users can tailor their usage to their needs.

Manuel Odendahl, a seasoned engineer, introduces "vibe coding" during the Agentic Engineering Sessions, showcasing his method of using multiple AI models simultaneously as specialized tools for software development. This approach diverges from traditional prompt engineering by focusing less on the immediate output of the models and more on the overall process, utilizing YAML DSLs to bridge natural language and executable code. Odendahl illustrates this technique by constructing a Doodle calendar clone in real-time, highlighting the importance of managing expectations, learning from AI failures, and guiding models without micromanagement. He emphasizes that while vibe coding is not a complete replacement for traditional coding practices, it offers a unique modality akin to pair programming or whiteboard design sessions. Throughout the session, Odendahl shares insights into the evolving nature of AI tools and the skill required to effectively harness them, all while continuing to learn about the systems he interacts with, even without directly reading the code.

Zed Industries employs a unique approach to hiring, where developers can join the team not only through traditional job applications but also by making significant open-source contributions to their GitHub repository. This innovative pathway has led to the recruitment of developers like Junkui Zhang and Anthony Eid, who started by addressing key issues and adding valuable features to Zed's codebase. Junkui's contributions significantly advanced Zed's compatibility with Windows, while Anthony's work on the debugger feature exemplified the collaborative and hands-on culture at Zed. The company values diverse backgrounds and emphasizes the importance of passion, communication, and the ability to deliver quality work, encouraging potential contributors to engage in pair programming and become part of their dynamic team.

A recent issue in Zed's settings parsing led to a malfunction in the auto-update feature for versions v0.203.0 through v0.203.4 and v0.204.0, which has been corrected in versions v0.203.5 and v0.204.1. The problem arose from a change in code that mishandled settings files, causing the auto-update default to mistakenly be set to false if not specified by the user. Despite automated testing and team vigilance, the bug was overlooked before the stable release due to communication challenges. In response, the team prioritized notifying users to update manually and implemented a test to prevent future occurrences of this issue. Moving forward, the team plans to enhance their settings management and improve their internal processes to better catch potential regressions.

Developers have eagerly awaited the integration of Claude Code into Zed, and it's now available in public beta through the newly developed Agent Client Protocol (ACP), which facilitates seamless connectivity between AI agents and editors like Zed and Neovim. This integration allows Claude Code to function as a first-class citizen within Zed, enabling developers to track real-time edits, review changes across multiple files, and utilize custom workflows through slash commands. Zed's implementation of ACP ensures a standardized interface, allowing Claude Code to operate independently while enhancing user interaction via Zed’s interface. The Claude Code adapter has been open-sourced under the Apache license, encouraging broader adoption and collaboration within the developer community. This move highlights Zed's commitment to expanding its platform's versatility by supporting a variety of AI agents and inviting contributions to further develop the ACP ecosystem.