Let me be honest about where AI helps in evaluation — and where it doesn’t.
The point worth examining
It’s genuinely good at the mechanical layer: parsing every slide, pulling the metrics founders bury, flagging what’s missing, scoring the same way across a hundred decks at 2am without getting tired. That’s the work that burns people out and breeds inconsistency.
What it shouldn’t do is decide.
What gets missed
Marc Andreessen described early-stage investing as “a qualitative evaluation, not quantitative.” Conviction, taste, reading a founder — that stays human.
“AI should read every slide. It should never make the call.”
— ai.STARTUPJURY field note
The mistake isn’t using AI in evaluation. It’s asking it to have judgment it doesn’t have. Use it to make the first pass consistent, so your people spend their judgment where it actually matters.
The evaluation problem
Where would you draw that line — what would you let AI touch, and what stays human?