You are responsible for your agents

One of the things I implement when operating a business is to ask whoever is responsible for a goal to write their own standard operating procedure.

Even if they are new to the business.

I can show them what needed to be done. We can both agree on what good work mean. I can show them the last experiment that worked.

But that’s for them to build on, not just adopt.

I now use a similar framework when using AI tools - for research or when playing around.

I don’t write the prompt myself. I describe what I want to get done, what the best case scenarios look like, and tell it to produce the prompt to make this happen.

Not surprising, it spills out a more detailed prompt for what i want. If I smell a faulty assumption in the prompt, I call it out and ask that it recreates the prompt.

Let whoever is responsible for the work take 100% responsibility.

For that phase, the AI is responsible. For the outcome, I am responsible.