80 percent is good enough
People ask me if they should worry that the agent's work is only 80 percent right. I would not. At the start of anything, 80 percent is good enough. When it drifts, you say stop, check it, replan it, and do it again.
- Where
- In the web chat
- Size
- A quick one
- Short link
- piy.show/s103
The worry at the start
People ask me the same thing when they start. The agent's work is only 80 percent right. Should I worry? I would not worry about that. At the start of anything, 80 percent is good enough, and especially if it is you doing it.
Here is why. The first version is not the product. It is the thing you react to. A blank page gives you nothing to react to. An 80 percent page gives you a list of what is wrong, and that list is the real plan.
So ask for the rough version first, and ask the agent to tell you where it guessed. Then you check the guesses, not the whole thing.
Why is 80 percent enough at the start? Prompt it yourself
Build the first version now. Tell me what you are sure about and what you guessed, so I know which part to check.
When it starts to drift, say stop
At some point the agent starts to hallucinate, or it wanders off. That is normal. You can always say: stop. Let's check it out.
Those are the words, and they are enough. Stop ends the drift. Check it out makes the agent show you what it did, step by step, so the two of you look at the same thing.
Do not argue with a drifting agent. Stop it, look, and then decide.
What do I say when the agent goes wrong? Prompt it yourself
Stop. Show me what you did so far, step by step, and tell me where you are not sure.
I can help you build yours
I build things like this with an AI agent, on my own computer. In a private session I help you do the same with your idea. Or bring it to the show, free.
Replan it, and do it again
Then: let's replan it, and let's do it again. Or, if the plan was fine, let's fix your issue.
A plan is cheap. Redoing from a better plan is faster than patching a wrong result ten times. The agent does not mind starting again, so the only cost is your decision to say so.
Replan when the shape is wrong. Fix when the shape is right and one thing broke.
When do I replan, and when do I fix? Prompt it yourself
Replan this from the start with what we know now, then do it again. If the plan was right, fix only the issue.
Start with your 80 percent
Pick one thing you have been putting off. Ask for the 80 percent version today. Read it, say what is wrong, and let the agent do it again. That is the whole loop, and it is a good loop.
What do I do next? Prompt it yourself
Give me an 80 percent version of this today. I will tell you what to fix.
Prompty