Show HN: Whiteboard (YC W26) – An open-source IDE for thoughtful software design
TL;DR
Whiteboard is an MIT-licensed desktop app built on a vendored Code OSS. In it, a coding agent such as Claude Code or Codex draws diagrams of its work that link straight to the code. Whiteboard adds a semantic diff viewer and a log of the decisions the agent made on its own. The goal is to let developers review agent-written changes without piling up "cognitive debt".
The article
- Four former tech leads built Whiteboard after they merged agent-written PRs without understanding them. Their own codebase became hard to work on. They call this "cognitive debt", after Geoffrey Litt.
- The agent gets an SDK and MCP server to draw on an in-app canvas. Clicking a sequence diagram, an entity relationship diagram or a quote from the agent's trace jumps to the code. Code navigation has VS Code keybindings and LSP support.
- A Rust, AST-aware diff viewer cuts noise by default. It summarizes large new functions as pseudocode and collapses tests and documentation changes. WASM plugins can change those rules.
- A decision log links the agent's own traces to the canvas. A reviewer can see how each stated requirement was implemented. The log also shows which choices the agent made without being asked.
- The founders vendor Code OSS instead of maintaining patches, because agents handle patch sets badly. Much of stock VS Code is dead weight to them; they say about 45% is Copilot. Whiteboard cannot edit files yet, and shared reviews don't update after sharing.
- The desktop app stays free and self-hostable. The founders plan to charge companies for a hosted, multiplayer version with trace storage and team reviews.
Why it matters
Developers who review a lot of agent-written code should read the README for the example prompts and the list of known limitations. The demo video shows the canvas-to-code navigation better than any description of it.
The discussion
A warm launch thread: the founders answered almost every comment, and several readers had built similar tools for themselves. Most of the argument was about the name, and whether a diagram canvas earns its own app over Markdown plus Mermaid or C4. A third question was whether humans should still review code or only plans.
An IDE that can't edit
The most repeated objection was the word "IDE" for an app whose known limitations say it cannot edit files. The founders said most people now use VS Code only to browse and review code. They held editing back because it pulls Whiteboard toward an agent orchestrator, and mid-thread they renamed it a "canvas" on GitHub and the website. One commenter defended the label as a frame of reference people would actually click on.
Seems more like IDE is "I Don't Edit" in this case
we ultimately settled on the term 'IDE' because we've found that most people use vscode to review code nowadays, almost exclusively (so it makes it a bit easier to draw the comp in your mind).
A product, or one feature?
Skeptics asked why this needs its own app when an agent can write Mermaid into Markdown, or LikeC4 and PlantUML's C4 library already exist. One called it "1 feature (a MCP with a GUI attached to it)". The founders answered that static tools lack go-to-definition from a node, guided walkthroughs and a real diff viewer, and plan an MCP UI version. A LikeC4 user replied that LikeC4 already has interactivity and walkthroughs, so a code view is a single feature.
I don't know how much better this is compared to some simple custom Mermaid format, which the agent can write as artifact files and present to users. Zooming out, this app seems like 1 feature (a MCP with a GUI attached to it) rather than an entire product.
My issue is that those tools are mainly built to render static diagrams (output to svg, png, etc.) … I think a lot of the value of whiteboard comes from the interactivity -- being able to click on a node to see the underlying code, being able to go through a guided walkthrough of a given flow, etc.
Review the plan or the code?
One commenter said agents plus automated review are already reliable enough that humans only need to review plans. A founder disagreed: key tradeoffs only show up once code exists, so writing code is a cheap way to improve the spec. A game developer said weeks of plan-only review left 50K lines of bloat and constant bugs. Another said agents get tunnel vision on code they just wrote, so restarting beats revising.
I have (like all of us I assume) tried moving forward over weeks of not code reviewing and only plan reviewing, and it's amazing how badly things fell apart, I ended up having to reset weeks of work and ended up knocking out like 50K loc for the exact same features and zero bugs
once the LLM has locked down an implementation any reworks I try to do gets it really tunnelvisioned on the current implementation, treating it like the truth even though it JUST wrote it
Diagrams that drift or invent things
Several readers worried the diagrams become one more source of truth that goes stale, or that the agent draws things the code doesn't do. One spotted a transition in the demo GIF that wasn't in the diff. A founder called that GIF a design mockup, swapped in a real session, and said real diagrams are tied to code. A cofounder conceded the diagrams are ephemeral, and some readers were happy to use them as a throwaway memory aid for their own agent-built code.
I'm pretty wary of LLM development tools hallucinating and wasting my time, is that whats happening in the lease broker example?
When using an agent for development, I find that I REALLY miss being able to "see" the shape of the code in my head while I work, and I find that I get bombarded with so much information that I cannot keep track of what the code looks like.
Platforms, size and agents
Many replies were requests: Windows (shipped during the thread), AUR, Nix, .deb, AppImage, Copilot CLI support, light mode. One reader balked at a 736 MB macOS download. A founder blamed the VS Code base, noted the diff viewer alone is 138 MB, and said a native rewrite is planned. A Codex warning about an "authoring server" turned out to mean the local stdio MCP server.
we made the choice to build on top of vscode for the mvp so that the code navigation experience would be normal / seamless (and hopefully, devoid of slop).
Consensus
Reviewing agent-written code is a real and growing pain. Many commenters had already hacked together their own explainers or walkthrough tools for it. Linking diagrams to code and the pseudocode-summarizing diff viewer got the most praise.
Dissent
A minority saw Whiteboard as a narrow feature that Markdown, Mermaid, C4 or a VS Code extension could cover. A few read it as one more step toward letting agents write code people no longer understand. A founder argued it aims at the opposite.