Inspired by tools like "Browser MCP", I wanted a way for LLMs to see my CLI/TUI applications during testing/debugging to help me troubleshoot issues quickly. Yes, LLMs can already spawn terminals on their own, however this allows both you and the LLM to interact concurrently.
There are other creative uses for it... like I can get Claude Code to use the full Gemini or Codex CLIs and ask for help on a problem. Others have pointing out achieving similar results via tmux, however my approach is simpler without that extra layer.
What you’re doing is coming…Microsoft is already working on this problem—the issue is trust. I only see this working as an open source initiative where people can inspect the code, run their own servers and provide their own keys. Maybe do open core in which you upsell certain reports or other features.
You’re implying that if someone uses AI to write something, the person doesn’t then read it/iterate on it to ensure correctness. Serious “get off my lawn” vibes here.
A journalist digging for details before writing a story is what one expects. The real question is how reputable is this journalist and outlet? What's the risk here? That they'll quit their job and go copy your thing instead? If answering that question puts your company at risk, then the issue may be the level of value you provide on top of whatever tools/frameworks you used.
I have 108k unread emails on Gmail. “Analysis” as I read them isn’t my pain-point. Surfacing the 2-3 important emails I get daily so that I don’t miss something important among all the noise is really all I want from AI.
Everyone is focused on helping me write emails or summarize emails; things I frankly do not care about.
That said, I like the idea of a copilot for the browser, it just requires a great deal of trust in your company since I’m giving you access to my entire life.
Patricia Arquette’s car was most disorienting to me. This woman was a high level manager/exec at a very successful company, surely she can afford more than a rabbit.
I’ve been considering this to get media on my Surface tablet. There’s no “legitimate” way of having offline content on a Windows tablet which baffles me. Netflix, Prime, Disney… none of them will allow you to download content for offline viewing on Windows.
I started building a client too but the fact that one needs to handle a user’s creds directly just felt off to me. Don’t understand how this is an accepted practice.
An idea I’m working on is allowing the integration of APIs by writing a prompt. I’d have a large number of APIs of different types and I’d discover the best endpoint and execute it and return the results. First time your code runs would “compile” the prompt, so subsequent execution would run O1.
This could be used by a human or an AI agent.
let temp = Get(“current temperature in NYC in F”)
Simple scenarios like these are easy, but looking at more sophisticated stuff as well.