On programming with Agents (from Zed blog)
Yesterday I was quite interested in an article by Zed about working with agents. I work with them in a very similar way.
The main idea: LLMs automate writing, not thinking.
The three rules that are described there:
Use the agent only for tasks that you can already do yourself
Stay in the loop – watch what it does, don't let it go on autopilot
Review, review, review
I have a similar situation with Claude Code in Zed. I delegate the "dirty work", but the final decision and review is always up to me.
More details: https://zed.dev/blog/on-programming-with-agents
0 Comments