The biggest trap I see with AI in estimating is overusing it. And that's from someone who genuinely loves Claude.
You've probably heard the line that using an AI model is like having a room full of brilliant graduates. It's not quite right. These graduates have no common sense, they have amnesia, and they don't learn anything unless you explicitly tell them to.
I ran into this on a battery project. I had a detailed commissioning estimate and asked Claude to check nothing was missing. It came back with 50 recommendations — most of them already covered in my construction cost or built into my overheads. Any estimator who understood the project would have known that instantly.
So the way I use it now is simple. Narrowly define the task. Give it the background it needs. Define the process. Specify the output. AI does the heavy lifting, you do the steering.
Which means you still have to understand the estimating process yourself.
Let me know if you have any questions.