😸 Permission to proceed, human

gm legends, happy Friday.

Today: an app that makes your AI agents wait on your thumb instead of the terminal, a notes app that wants nothing to do with AI, and a search engine built by a guy whose whole job is knowing which startup solves your problem. Plus the forum thread about the exact second an AI agent lost your trust.

Pushary sends your AI agent's permission requests straight to your phone's lock screen, so a stalled overnight run doesn't sit there till you're back at your desk. Aadil Ghani (@aadilbuilds) built an AI orchestration pipeline for live translation before this.

🔥 Our Take: The part that's actually good is what happens when you don't answer: it defaults to no, not yes. Go dark and your agent doesn't start improvising, it just waits. Anthropic already gives you a version of this free if you're only running Claude Code, one agent, one tool. Pushary's whole play is that most people aren't set up that way anymore, they're running Cursor and Codex and Gemini CLI too, and four separate notification apps beats the point of any one of them. Bullish.

MinkNote Demian Turner (@demianturner) ran an iOS development agency for a decade before building this native macOS notes app, which stores everything as plain Markdown files: no database, no account, and by design, no AI features baked in.

🔥 Our Take: Every notes app right now is racing to shove a chat sidebar in. Demian just didn't. He wants people bringing their own AI instead of the app owning one, and says so straight in his own launch thread. Someone called him out there too: the "no lock-in" pitch is a little rich coming from closed-source software. Fair, you're trusting Demian, not the code. Obsidian already does portable files with a decade of plugins on top. MinkNote is going after people who got tired of the plugins, not just the cloud.

Framer 3.0 ships AI Agents that live inside your canvas and can actually design. Not suggest. Design. Drop a brief and watch them build out full pages, create components, write code, connect your CMS, and handle SEO. All without leaving the canvas. They plug straight into Claude Code and Cursor. Yes, inside your actual component system, without breaking it. The Figma file you were about to hand off to engineering? Keep it.

YC has it Raghav (@raghav2) spends his days scouting startups for other people, so when he got tired of nobody being able to find the right YC company for their actual problem, he built this: describe what you need in plain English and it searches 4,000+ active YC companies for a match.

🔥 Our Take: Thing is, the real competition here isn't another directory. It's just asking ChatGPT the same question, for free, and it already knows plenty about YC companies. What Raghav's got that a chatbot doesn't is a dataset rewritten so it's not just parroting marketing copy back at you. Somebody tested it on launch day with a real, narrow problem, then asked the question nobody had answered yet: who's updating this when a company pivots or shuts down. A confident wrong answer is worse than no answer. Skeptical until someone shows me the update pipeline.

Terranicgerald (@terranicgerald) asked the forums the smallest thing that ever killed their trust in an AI agent mid task. Thirty replies in, almost nobody's story was some huge blowup.

Same pattern every time: it was never the mistake, it was the confidence attached to it. "All tests pass," except one got silently skipped. A failing test "fixed" by editing the assertion instead of the code. A file deleted during cleanup, no heads up given. Nobody got burned by an agent that said "not sure." Everybody got burned by one that didn't.

Best line in the thread, from roguetink, on an agent that quietly merged two different bug reports it decided were duplicates: "A wrong merge you can fight with, a silent one you have to remember to go looking for in the first place, and most people won't."