For the past few months, most of my energy has gone into something new. I am building a project around AI agents — not an app with a chatbot bolted on, but a project where the agent does the actual work.
Two tools sit at the center of my daily workflow right now: WorkBuddy, an agent that lives on my machine and can read files, run commands, and drive a browser, and Gemini, which I use for research, writing, and multimodal analysis. Together they have changed how fast an idea becomes something real.
From tools to teammates
I have used AI as a search engine and a writing helper for years. That was useful, but passive: I asked, it answered, and I still did all the doing. Agents are different. You give them a goal — “log into this site, check what changed, and write it up” — and they plan and execute the steps themselves. The first time you watch an agent fix a bug, publish a post, or clean up a spreadsheet while you make coffee, it is hard to go back.
What surprised me
- Speed of iteration. Ideas that used to sit on a “someday” list for weeks now become a working prototype in an afternoon.
- Delegation is a skill. A vague instruction gets a vague result. Writing a good task description for an agent is 80% of the work — and it is the same muscle as managing people.
- Trust is gradual. I started by letting agents do safe, internal things: reading, organizing, drafting. Only after they had earned it did I let them touch anything that leaves my desk.
What I am building
The new project takes this way of working and turns it into something others can use: less clicking through software, more describing what you want done. It is early — the foundations are the interesting part right now — and I will share progress here as it takes shape.
If you are experimenting with agents yourself, I would love to compare notes. Reach out any time.