Personal Software?
Back in ~2015 we were shutting down my first startup (long story), and I was trying to figure out what I wanted to work on next.
Hardware was becoming powerful and cheap, so I started hacking on an OpenWRT-based router/server, with the goal of having an “always-on” hub that could both connect my home internally as well as provide an interface for external access.
The project died when commercialization extracted all the fun.
The hardware was mostly there but finicky, however the software landscape was quite rough.
I still think it could have worked — in some sense it did.
10+ years later, however, and personal software is here to stay.
Why now?
Large language models have reduced the cost of writing software to almost nothing*.
I’m writing this blog, using a home-grown CMS built on top of Astro, that I built with Claude Fable 15 minutes ago.
The existing solutions didn’t fit my needs, and building the markdown editor -> deploy pipeline was faster than importing and dealing with the contraints of an existing system.
*Writing software is cheap, building it still takes time and preference.
Aren’t agents the future?
Yes and no.
I like to think of anything agentic in terms of humans.
Do I want a human processing every credit card transaction?
No
Do I want a human updating the dashboard for my product each day?
No
Do I want a human telling me my tire pressure is low?
No (well yes, if the tpms sensor is broken!)
Agents are great, but I believe the line will blur between agent and software more than we can perceive today.
These days, when someone says agent, they usually mean an LLM in a loop with a text or voice interface.
Agents of the future will coexist with software.
Machines will finally work how we always expected them to.
Computers are agents. Agents are computers.
Ok, so personal agents/assistants?
A little bit of column A, little bit of column B.
When most people think of “personal AI,” they think of an Alfred or Jeeves-like digital EA that parallels your life much like a human might.
That’s cool, but I think the future probably looks more like a personal team of software engineers.
I could ask my EA to go get mangoes from the store, or my personal dev team could suggest acquiring a unique mango varietal available in Fremont (no association), after reviewing the list of crowd favorites from last year:
...next-year recommendations (more Zill citrus, identify Kathy, skip Tierno).