Filed under:
I consider myself lucky to have access to an AI coding agent subscription at work.
I've read and seen crazy things being done with AI nowadays, mostly people becoming rich. As for me, well, I'm just living as usual, can't do that, not interested in doing that.
I've seen posts about how to optimize coding agents, but I felt like most of them were written by agents. (Chuckle.) It made it hard for me to decide whether I should read them or not.
I use AI for blogging, but only for simple grammar fixes so that, at the very least, my posts have the correct grammar for depicting the past or present. My prompt is very simple:
grammar fix, keep character: <post content goes here>Oh right, back to coding agents.
A few lessons I've learned so far are that it all comes down to how you structure the information.
For one, it's better to isolate your agent's memory into its own resource folder and build upon it over time rather than keep asking the agent to explore the source code directly. That way, the agent goes to the context first and the source code second. It improves accuracy and lets you continue across sessions.
Second, each session should be well-scoped, whether it's a list of skills that you'll use or a set of first-class rules that gate certain decisions for that session.
To achieve this, all you have to do is prepare a single command that, once called, brings in all of this tightly packed, well-defined setup that you can reuse whenever you're working on a particular project.
**
That's all I want to share for now.
https://www.blogger.com/rpc_relay.html https://www.blogger.com/comment/frame/8166404610182826392?po=7719598596166398099&hl=en&saa=85391&origin=https://vanillawebdev.blogspot.com&skin=contempo https://vanillawebdev.blogspot.com/2026/07/coding-agent-scoping.html#comments true ["agent"]