@yibie: Every programmer should hand-write an agent. It only takes about 50 lines of code, it's fun, and it will surprise you. But to get the most out of it, do these two things: First, write it from scratch. Open a blank text file and type every line yourself. Don't use any AI, not even autocomplete. Second, rely only on the standard library documentation…

Every programmer should write their own agent.

It only takes 50 lines of code, is a lot of fun, and will surprise you. But to get the most out of it, do these two things:

First, write it from scratch. Open a blank text file and type out every line of code yourself. Don’t use any AI, not even autocomplete.

Second, rely only on standard library docs and API docs (e.g., Anthropic API).

You could have Claude help you write it, but the value of building it yourself is this: it builds a strong mental model of how an agent works — not a rough “LLM in a loop with tool calls,” but every detail.

When you do everything by hand, you won’t accidentally skip important parts you don’t yet understand, and you’ll see firsthand how quickly new behaviors emerge.

My first agent had only one tool — “add two numbers” — but by repeatedly calling it, it could already do multiplication.

Once it works, there are dozens of ways to play: add a bash tool (remember to do it in a sandbox), switch models (can the worst model still produce good results?), or build a simple memory system.

— Nick Bergson-Shilcock, Recurse Center

Original: https://x.com/nicholasbs/status/2070529196351238252…

#AI #Agent

Nick Bergson-Shilcock (@nicholasbs): Every programmer should write their own agent.

It’s fun, only takes 50 lines of code, and will surprise you. But to get the most from the experience, do these two things.

First, hand-code it from scratch. Start with a blank text file and type out every line of code yourself;