In the previous note we looked at which tasks are actually worth automating. The next question is: how do you actually work with AI on those tasks?

Observing how companies use AI, two very different approaches emerge:

  • The first is using it as an assistant: you query it now and then, ask for one small piece at a time, copy the answer. It's the most common approach, practically an improved search engine.
  • The second is treating it as a colleague: you hand it an entire task, with the project context, the instructions and the criteria the result will be judged by.

The difference in results is clear. You ask an assistant to "write this function"; you tell a colleague to "implement this feature in the project". The first returns a fragment to adapt, the second returns work to review!

For the second approach to work, three things are needed: clear instructions, complete context, and a final human review - the same you would apply to a collaborator's work.

When one of the three is missing, results turn generic and the time saved is lost in corrections. It's not a limitation of the tool: it's the method it's used with.

Most companies today exploit a fraction of the potential of the AI they already pay for, because they only use it the first way.

Those who learn to delegate to AI the way they delegate to a collaborator will gain extra productive capacity. Those who keep querying it like a search engine will remain convinced that AI is overrated and the results will prove them right!

Luca Pernini - Script Solutions .IT