gm legends
Four today with nothing much in common: talking at your coding agent instead of typing at it, an engine that started out making video game characters talk, and Block made AI creatures.
gm legends
Four today with nothing much in common: talking at your coding agent instead of typing at it, an engine that started out making video game characters talk, and Block made AI creatures.
| • |
Aloud turns you talking at your screen into a task list for your coding agent. Hit record, walk through your app pointing at what's wrong, director's commentary style. It cleans up how people actually talk, the "move this over there" and the changing your mind halfway, and works out which frame you meant. Out comes a set of tasks for Claude Code or Cursor. The slow part of building with agents is the brief now, and Wojciech Dobry was losing five minutes typing what he could say in thirty seconds. |
| • |
Record for Agents hands your coding agent the context you would otherwise have typed out. Narrate the problem on screen and Loom captures the clicks, the pages you visited and what you said about them, then writes the lot up as a structured prompt. You review and edit that before it goes anywhere, which is the part most bug reports skip. From there the agent can draft the ticket, write the code or carry out the task itself. It is built into the new Loom for Mac and also runs in Chrome, and it is in open beta for Business + AI and Enterprise. Check out the demo. (sponsored) |
| • |
Berd is a desktop app built by Block (Square, CashApp, etc) for running AI agents on your own files, and instead of a chat window you get an infinite canvas with each agent sitting on it as a little creature called a Gloopy, which is the maker's word and not mine. Your files, skills and sessions live in the same space, history stays on your machine, and it works with whatever model you already use. It runs on Goose, the agent framework Block open sourced last year, now past 53,000 GitHub stars. |
| • |
NobodyWho is an inference engine that runs AI models on the device: any open-weight model, images and audio in, Whisper for transcription, speech out, no API keys and no bill per token. It began life as a Godot plugin so indie game NPCs could hold a conversation instead of reciting the same three lines forever, and grew into an engine that also runs on Swift, Kotlin, Flutter, React Native and Python, with demo apps down to the Apple Watch. The grammar is built from your function signature, so a small local model can't hand back malformed JSON.
|
