DEV.to developer Isaac Natarajan is dogfooding his own agentic AI project hard, and it shows. In a post published August 3, he details 'my-assistant,' an AI agent built to solve one deceptively simple problem: finding his own files. V1 is live, real, and something he uses every single day. It's not polished and it's not finished — but it works, which puts it ahead of half the agent demos floating around right now.

The Problem Nobody Fixes

File search on a personal machine is broken in a way that stays invisible until you slam into it. Your OS indexer can match filenames and maybe content keywords, but it has zero understanding of intent — 'the invoice from March,' 'that doc about the thing,' the config file you renamed three times. Natarajan started learning agentic AI the standard route: by watching a course. But instead of stopping at toy examples, he turned around and pointed an agent straight at his own mess.

V1 Is a Daily Driver, Not a Demo

The headline here is honesty about status. Natarajan explicitly frames this as 'a real, working assistant I use daily,' not a finished product — and that distinction matters for anyone building personal agents. A tool you actually trust to dig through your own files has cleared the hardest bar in agent development: you keep using it after the novelty wears off. Most hobbyist agents die at week two; this one survived into production-of-one.

V2 Throws the Doors Open

He's already planning a second version and openly soliciting input from readers — 'I'd genuinely love your ideas on it,' he writes, teasing more detail at the end of the post. That open-loop design is smart for personal tooling, where an agent's utility depends entirely on how well it adapts to one person's chaotic file habits rather than some idealized benchmark set.

Key Takeaways

  • Natarajan built a working AI agent for personal file retrieval that he uses daily — not just a demo.
  • It addresses a real gap: OS indexers lack the ability to understand intent behind vague search queries.
  • He's openly inviting community ideas for V2, treating his own project as an open-loop experiment in personal tooling.

The Bottom Line

The takeaway isn't that agents can search files — plenty of products claim that. It's that Natarajan built something he actually lives with, then had the guts to ship V1 in public and ask for help making V2 better. More agent builders should be this honest about what 'done' looks like.