Execution Got Cheap. Judgment Didn't.

Duration: NaN:NaN:NaN

Speaker: Pragya Bhardwaj

Type: CLS - Lightning Talk (Open Tech Summit)

Expertise:

Event: Devcon

Date: Nov 2026

My first two roles were data operations and frontend development. Both are roles people now say AI has made cheap. I moved into product before that conversation started, and the part of the job that still matters is the part nobody can prompt their way to. Earlier this year I had to choose between three external data sources for a platform I own. All three were buildable - the builders were ready and could have shipped any of them. What made it hard were questions no tool answers: which one gives users value by default, which one returns mediocre answers but solid citations, and what each one actually costs to run and maintain. I'll trace where that habit came from - Google Summer of Code, where I proposed things in public and got told no, with reasons, by people with more context than me. And I'll defend the narrow claim: not that AI can't reason, but that someone always has to sign off and be answerable. Accountability doesn't get automated. You'll leave with those three questions and where to use them.