Sign in Subscribe

What's the most practical and productive way to use AI Agents today?

Nguyên Phan

• 2 min read
The best advise of mine is to keep constantly exploring new methods, new things everyday. That's really the only way to keep up and ahead.

So, up to date, what's my personal method?

I use combination of Claude Code and Codex. There are many articles online that says "use this for this and use that for that" but I find it's not reliable. Because the frontier AI providers are competing day by day, and even within the same model, same everything, but different conversation/session produces different quality and result. So I just switch between the two of them, have them guard each other, argue, cooperate to produce the result that I want.

For Claude Code alone, I encourage you to use whatever mode is most capable for complex jobs. For now it's Ultra mode which is xhigh effort + /workflows. You can tell your Claude Code agent to "use workflows" manually too.

I setup an ubuntu machine where I control everything on it, So the blast radius is minimal when compromised. I use claude –dangerously-skip-permissions and codex -s danger-full-access for better and faster result.

So until now, you've assumed that I only use those agents for development tasks? No, this is where you must know, you can do so far more tasks than just development, you can do all kind of tasks like:

  • Edit videos, working on creative jobs in general.
  • Manage servers, personal devices, make your devices are secured.
  • Research something.
  • Legal works, Paperwork, Office works.
  • All sort of things you can imagine...
You must know where you files are. I usually use the agents using CLI, and I always keep files and documents in a directory not relying on agents' memory or planning. I always use git and GitHub for every single project no matter the scale. That way, no agents can destroy your work in a meaningful way.

But still, I explore new more efficient and productive ways to work everyday. Oh, one last thing, please don't mind the price you have to pay, AI bills will pay themselves, frankly, those big AI Providers are really not making money on selling subscriptions to individuals. You are in for a good deal. I will write a post about privacy with local LLMs in the near future.