I use AI tools every day, and I do not let them decide anything. My rule is short. The AI suggests, a person decides, and something other than the AI checks the result.
What AI is good at
Covering ground, offering options, finding faults and writing first drafts. I often ask two different AI models the same question without showing either one the other’s answer. Where they disagree is where I look hardest. When they agree I am more confident, but I still check.
What it is bad at
Being right when nobody is checking. A smooth answer is easy to mistake for a correct one. I keep a running list of things AI tools have told me that turned out to be false. Each one becomes a test or a check, so the same mistake cannot get through twice.
Use a script when the answer never changes
Some tasks have one correct result every time: renaming files, comparing two lists, checking links. Those do not need AI. A short script is cheaper and faster, and it does the same thing tomorrow. Save AI for work that needs judgment, and keep a person responsible for the result.
Match the tool to the task
- Work that needs real judgment, such as design, security and settling disagreements: the strongest AI model, with the closest human review.
- Routine work that follows an agreed plan: a cheaper model is fine.
- Mechanical checks: no AI at all.
Say when you used it
Tell people when AI was involved. Keep a history of who changed what. And never let the tool that did the work be the one that approves it.
This is how I approach AI, automation and process consulting and my own development work.