Give the Tool, Not Only the Prompt
For some months I managed my readings in a spreadsheet: books I own, books I already read, books I want to buy, buy priority, and so on. Every time I got a recommendation or finished a book, I had to open the sheet and update a specific row manually. I started thinking about handling those updates with natural language, using AI. Simple prompts like "list me all the books I have in the finances category" or "add these 6 books to the buy priority list."
Then I remembered a good blog post called AI agents don't need your developer tools, and one quote caught my attention:
So if agents don't need our tools and they don't need our frameworks, what do they actually need?
The honest answer is mostly boring: well-documented APIs with stable contracts, reliable file systems, clean version control semantics, structured I/O, and sandboxing. The kind of infrastructure that has existed in some form for decades.
Based on that, I decided to create a simple CLI so agents could manage my book lists. I built books-cli.
The power of a CLI
AI agents are very good at running bash commands, so a CLI is cheaper and easier to use. Along with the CLI, I wrote simple docs that teach the AI how to use it: how to edit a book, update fields, list, paginate, filter, and search by term. For storage I use a plain SQLite file, so I can back it up easily on my machine.
I've been testing this for some days and it works very well. I get a better picture of my books with simple prompts like "list me all the books that I want to sell" (because that's a field in the database) or "based on the book I'm currently reading, list books from my buy priority list with the same theme."
The same idea applies elsewhere. I wanted a fast overview of GitHub items assigned to me across repositories. I didn't build a complex app or a long prompt to merge the data. I wrote a simple bash script that uses GitHub CLI to fetch the items.
Agents are fast and smart, that's obvious, but they need the right tools and the right instructions.
Before you automate your next task, think about it. Sometimes a prompt is not good enough. You need to provide a tool. Teach your agents how to lint files, run tests, fetch items from GitHub, and so on. Maybe a bash script or a small CLI with concise docs is enough to improve your work.
P.S. This entire post was written using the Cursor feature to record voice, so basically what you read, I was dictating to the AI. If you are not a native English speaker, it's a very good option to train your speaking. If you aren't a Cursor user yet, you can try it through my referral link.