Show HN: Whiteboard (YC W26) – An open-source IDE for thoughtful software design
Whiteboard is an open-source desktop application designed to bridge the gap between human developers and AI agents in software design. It offers visual diagrams, semantic diffs, and decision logs to help make sense of agent-generated code. This project resonates with the HN community grappling with the new challenges and 'cognitive debt' introduced by AI-driven development workflows.
The Lowdown
Whiteboard, a new open-source desktop app from YC W26, aims to solve the growing 'cognitive debt' developers experience when working with AI coding agents. Born from its founders' own struggles, the application provides a collaborative workspace where humans and AI agents can architect software together, making the AI's output more transparent and understandable.
- Agent-Human Collaboration: Whiteboard integrates with existing AI coding tools (e.g., Claude Code) to allow agents to draw directly on an in-app canvas, explaining their work through diagrams and traces.
- VSCode Foundation: Built on CodeOSS, it offers familiar code navigation and LSP support, enabling direct jumps from visual diagrams (like sequence or ERDs) to the corresponding underlying code.
- Semantic Diff Viewer: A Rust-powered, AST-aware diff viewer reduces noise in code reviews by summarizing large changes, collapsing unit tests, and hiding documentation updates, all customizable via a WASM-based plugin system.
- Decision Log: It tracks and visualizes an agent's autonomous decisions, linking them back to initial requirements to provide clarity on implementation choices.
- Practical Use Cases: Companies like Salesforce and Modal are already using Whiteboard for reviewing AI-generated prototypes and escalating complex changes that require human judgment.
- Open Source & Future: The desktop app is MIT-licensed and self-hostable, with plans for a hosted web version offering team-focused features like trajectory storage and multiplayer reviews.
By providing tools to visualize, interpret, and review AI-generated code, Whiteboard endeavors to make the burgeoning field of agentic coding more manageable and comprehensible for human developers.
The Gossip
Diagram Dilemmas & AI Accuracy
Commenters raised immediate concerns about the accuracy and potential hallucinations of AI-generated diagrams, especially if they are not directly linked to code. The author clarified that while an initial demo GIF was 'notional', actual diagrams within Whiteboard are tightly coupled with the code, aiming to prevent such inaccuracies.
Nomenclature Nuances & Product Positioning
The use of 'IDE' in the title sparked debate, as Whiteboard currently lacks editing capabilities. Users suggested alternative descriptors like 'brainstorming tool' or 'canvas', questioning its differentiation from existing IDEs or if it should be a VSCode extension. The author acknowledged the 'misnomer' and is exploring better terminology, emphasizing Whiteboard as a complementary tool for understanding rather than a full editing environment.
Agentic Architecture & Workflow Evolution
Discussion centered on how Whiteboard fits into the evolving AI development workflow and its impact on cognitive load. Many agreed on the problem of managing AI agent traces and 'vibe-coded' systems. Commenters debated the future of 'plan -> code' workflows, with some suggesting a stronger focus on planning before implementation, while the author posited that code writing itself can be a 'cheap effort' to refine and make specifications more thorough.
Source of Truth Scrutiny
A significant concern was whether Whiteboard creates an 'N+1' source of truth that will inevitably become outdated, similar to traditional design documents. Critics questioned if software could truly solve this behavioral/structural problem. A co-founder responded that the hosted version aims to address this by capturing the entire evolutionary process from initial plan to agent implementation within a single canvas.