Learning from Others

One of the best ways to learn is by following experienced programmers who are shipping real projects with AI agents. Some of the most productive builders have surprisingly simple systems. They just talk to the model, let it do its thing, and don’t worry about extra slash commands, subagents, hooks, or skills. That simplicity is reassuring: you don’t need some ultra-complex system to be effective.

On Twitter and social media, you’ll see a lot of people really optimizing (or potentially over-optimizing) their own systems. That can feel daunting. But that’s also part of the beauty: it’s a completely customizable system. You can make it work however you’d like. You can have a plan mode that runs for twenty minutes with a custom slash command, or you can just talk to the model directly.

Another valuable practice is looking at other engineers’ open source software. Clone it, use it, try to improve it, or just take parts of it and make it your own. See a YouTube summarizer project? Take it, strip out the parts you don’t need, keep it as a CLI, and now you’ve got a tool you can use anywhere.

Reading posts where experienced developers explain their preferences (like choosing CLIs over MCPs) can give you the nudge to dive deeper into bash and command-line tools yourself.