Join Nostr
2026-11-06 09:16:18 UTC

gary_woodfine on Nostr: Whenever someone says, "I've tried AI and I found it unreliable and it just doesn't ...

Whenever someone says, "I've tried AI and I found it unreliable and it just doesn't work for me"

I dive in deeper, to find that when they say "try" they invariably mean just some chat interface and they're copying and pasting code or questions into it.

This is not "trying" AI, this is just chatting to it. There is a whole world of difference.

Here's an example for context and to get your head around.

When you're looking for Legal advice, you wouldn't just go up to first stranger in the street, give them a random, de-contextualised example of a situation and ask them to give you a definitive answer on how to deal with it.

No, you would first look for someone, trained within the legal profession, who has gained a broader understanding of the law, and then you would maybe drill down deeper to look for someone who has unique specialty within the law relating to your particular situation. So you wouldn't say ask a commercial property solicitor to represent you on a criminal case for instance.

No, you would look for someone with focused context on your specific situation.

The exact same principles apply to AI, you have to build the context, framework and confines for the model to work in. Give it as much context about your problem you are working on. You'll see the results are overwhelmingly better.

You don't do this with a chat prompt!

You do this by creating files and folders with as much detail, guardrails, context and data about your problem. You do this by giving your agent explicit operation guidelines.