Don't automate important tasks users do today; automate things they don't bother with today.
- Don't automate important tasks users do today; automate things they don't bother with today.[sk]
- If you automate important tasks they do themselves today, anything less than perfect could be below the bar of "good enough."
- If it fails even once, the user might learn they can't trust it.
- But there are likely some use cases that the user would get some value out of but it's not worth the effort for them today.
- If you can automate those, then you create new value that wasn't possible before.
- If you fail, the user is no worse off than they are today.
- All upside, no downside.
- This is similar to Ethan Mollick's frame on LLM quality thresholds of "best available human".
- Don't compare an LLM's quality to the expert in the field, compare its quality to what the actual human who would have done it instead is.
- This is often a much lower bar.