Claude/Gemini/Codex 10-100x faster with pandō (CAD for code)
Hi HN, I'm George Ciobanu (https://www.linkedin.com/in/georgeciobanunyc). I built pandō ('CAD for code') because I got tired of watching AI agents burn tokens, take forever, and still get it wrong. Here's (one reason) why this happens: AI agents read and edit code as if it's just text. But code is more, much more: it has structure, syntax, relationships, and meaning. Code is Data We built pandō around this insight. It indexes and persists your code's AST in a database, which gives AI agents instant search results, and enforces syntactically correct edits. Here's a full…
In plain words
Pandō is a development tool that helps AI agents edit code more efficiently by treating code as structured data rather than plain text. It indexes code's abstract syntax tree in a database, enabling faster searches and enforcing syntactically correct edits. The platform automatically snapshots changes for safety and prevents syntax errors through compiler-checked modifications. It is designed for developers using AI agents to accelerate code generation and editing tasks.
written from the facts on this page · September 2026
From the sources
In the maker’s words, at launch
Hi HN, I'm George Ciobanu (https://www.linkedin.com/in/georgeciobanunyc). I built pandō ('CAD for code') because I got tired of watching AI agents burn tokens, take forever, and still get it wrong. Here's (one reason) why this happens: AI agents read and edit code as if it's just text. But code is more, much more: it has structure, syntax, relationships, and meaning. Code is Data We built pandō around this insight. It indexes and persists your code's AST in a database, which gives AI agents instant search results, and enforces syntactically correct edits. Here's a full list of benefits that this different approach brings: 1. Correct Syntax - compiler-checked and gated edits (unless you force breakage) 2. Always Safe - auto snapshots after every change, so no matter what the agent does, 'Undo' is just a click away 3. 10-100x Faster - text-based tools must 1. read context around every match they find (match number scales with codebase size), 2. send that context to the LLM, 3. wait for inference, and finally 4. wait for the LLM to finish streaming the response. In contrast, pandō edits directly on your computer, in seconds. 4. Token Savings - >100× "token compression" for some operations (e.g. rename costs the same ~40 tokens whether the symbol has one or a thousand references). 5. Reduced Exposure - pandō operations send as much or less code to the LLM provider as text tools in order to achieve a goal; some operations send no code at all to the LLM — just the intent of the transform. Every edit is atomic, hash-verified, compiler-checked, and backed up by a snapshot. Download and install the Visual Studio Code extension (keep VSCode with the extension installed open) or download it from the marketplace(https://marketplace.visualstudio.com/items?itemName=humansof...), then connect from any MCP-compatible agent. Supported: Typescript, Javascript, Python, C/C++, C#. Coming soon: Java, Rust, Go, Clojure, R, Perl, Swift, Kotlin. https://getpando.ai feedback: george [at] human [dot] software
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
