I’m curious if we’re reaching a point where using an LLM as a general-purpose interpreter (one that has access to context, tooling, and function calling) can effectively replace the need to build custom tools. If an AI can call actions based on natural human language, why would I need to build a bespoke app?
For instance, I’ve been experimenting with this exact idea by using a Todoist MCP (Model Context Protocol) server and a Playwright MCP, I can essentially talk to any website, even if I don’t have API access. The AI can scrape the information I need locally, parse it, and then turn it into tasks automatically in Todoist. All of this is driven by simple, natural language queries.
Honestly, if I take it a step further and just pair that setup with a good voice model from OpenAI, ElevenLabs, or Deepgram, I could just talk to my system. I’d effectively have a walking, talking assistant that I pay next to nothing for, one that can manage and manipulate all of my personal tools. That’s really cool, and the 9-year-old hacker inside me would be in shock.
My next step is to see if I can create a robot that connects to my Claude Code instance. Could it take the context from my tools, feed it back to me like a human, and allow me to have a two-way conversation to get things done? This would be a system that works with all of my tools, not just the ones that OpenAI or other big companies decide to expose in their app-store-ified ecosystems (looks like someone has built a somewhat clunky/cool version that I’ve played with called Voice Mode).
On that note, I’d really appreciate the first developer who figures out how to run MCP servers directly on an iPhone. That would be a huge deal. We’ll probably end up putting them on the web first because servers are just easier for these models to handle, but being able to do it all locally on-device would be incredible. We would get to control the interaction and retain our own data.
I look forward to the future where we can run a local LLM and do all of these things privately and offline. We’ll each have our own personal robots working on our own computers, tasked with whatever we need them to do, so we can finally offload the difficulty of the mundane. Please, someone speed up with affordable laundry robots. Thank you 🙂