August 2025 Summaries

Zed has introduced the Agent Client Protocol (ACP) to enable seamless interaction with third-party agents within its integrated development environment (IDE), starting with the integration of Google's Gemini CLI as the initial implementation. This initiative aims to unbundle agent functionality from monolithic IDEs, akin to how the Language Server Protocol transformed language intelligence, allowing developers to switch between multiple agents without leaving their editor. ACP facilitates a structured communication framework using JSON-RPC endpoints, enhancing the capabilities of command-line agents like Gemini CLI by integrating them into a development environment designed for software creation. The protocol is open-source under the Apache license, encouraging adoption by other clients and editors, while ensuring user data privacy. Zed's collaboration with Google and the open-source community aims to build an ecosystem of agents tailored to diverse developer workflows, with the protocol also being adopted by other editors, such as Neovim. The project emphasizes openness and innovation, inviting contributions and feedback to further develop and refine ACP.

Jessie Frazelle, an experienced engineer, conducted an evaluation of various async AI agents to determine their effectiveness in real production environments, ultimately finding that most agents added more noise than value, often providing irrelevant comments or analyses. Her team, working with a diverse tech stack including Rust, C++, Python, and TypeScript, found that many tools, such as Copilot and Sourcery, frequently made unnecessary or inaccurate contributions to code reviews, sometimes even altering original comments or generating misleading flow diagrams. However, Jessie highlighted Graphite Diamond as a standout agent for its ability to identify significant bugs without contributing unnecessary commentary, which has become a trusted tool in her team's high-stakes projects. Despite her cautious approach to adopting new tools, Jessie appreciates the support these agents provide as a supplementary "second pair of eyes," especially in complex areas where errors can be costly. She concludes that while AI agents are not a replacement for human code review, the right tool, used effectively, can enhance a team's capabilities without overwhelming them with irrelevant information.

Nathan Sobo announces a $32M Series B funding led by Sequoia Capital, bringing the total funding to over $42M for the development of Zed, an innovative integrated development environment (IDE). The project aims to transform software collaboration by linking conversations directly to code, moving beyond traditional snapshot-based version control. Zed plans to introduce DeltaDB, an operation-level version control system using conflict-free replicated data types (CRDTs), to enable real-time collaboration with both human and AI agents, maintaining a continuous dialogue rather than relying on discrete commits. This approach allows for detailed, contextual interactions preserved within the codebase, offering a dynamic and navigable history of software evolution. As an open-source project with an optional paid service, Zed seeks to redefine developer collaboration by integrating comprehensive discussion and decision-making processes directly within the coding environment.

Zed, a software application, is progressing towards a Windows release, having transitioned from a one-person project to a full-time focus for four engineers over the past six weeks. Initially, Zed used a Vulkan-based rendering backend, which led to compatibility issues, prompting a shift to DirectX 11, reducing its memory footprint and ensuring broader compatibility with Windows 7 and later versions. The development process encountered challenges with glyph rasterization, leading to a switch from Direct2D to DirectWrite to work with RenderDoc for debugging. Moreover, Zed faced GPU memory allocation issues, which were mitigated by optimizing the multi-sample antialiasing (MSAA) process, enhancing performance across platforms. The auto-update system had to be restructured due to Windows-specific file system constraints, introducing an "auto update helper" for managing updates. Crash reporting was also adapted for Windows, using minidump files to handle symbolicating stack traces server-side. Future development will address key bindings, SSH remoting, WSL compatibility, extension path conventions, and performance monitoring. Users and developers interested in the Windows version can join the beta-testing group or contribute via GitHub, and the team is actively hiring.

Interviewing software engineers has provided insights into the key abilities that distinguish effective practitioners, particularly their capacity to build and maintain clear mental models throughout the software engineering loop. While large language models (LLMs) can assist in writing and updating code, they fall short in maintaining coherent mental models, often struggling with context omission, recency bias, and hallucination. Unlike human engineers, LLMs cannot yet manage the complexity of iterative problem-solving due to these limitations. Despite their current constraints, LLMs are valuable tools for generating code and synthesizing requirements, but human engineers remain crucial for ensuring accuracy and clarity in software development. At Zed, there is a belief in collaborative efforts between humans and AI, with humans taking the lead in driving the development process.

In a session exploring the future of Integrated Development Environments (IDEs), veteran engineer Steve Yegge and Zed co-founder Nathan Sobo shared their insights on the evolution needed for development tools in the AI era. They argue that traditional IDEs are becoming obsolete, stressing the necessity for these tools to facilitate collaboration and adapt to AI's growing role in software development. Both believe that current IDE architectures are inadequate, requiring a shift towards platforms that integrate AI as collaborative agents rather than mere tools. Yegge envisions a future where engineers supervise AI-driven processes, while Sobo highlights the need for high-quality platforms that can evolve over time. The discussion underscores the importance of reimagining IDEs as dynamic platforms capable of supporting autonomous coding agents, reflecting a broader shift towards agentic engineering where software quality, rather than production speed, becomes the primary focus.