Snowflake SVP says you can't spot your AI explorers

Snowflake's SVP of engineering, Vivek Raghunathan, says only about 5% of engineers become AI explorers, and seniority doesn't predict who, per the Stack Overflow blog. That's a problem for any manager building an AI rollout around a hero coder. Raghunathan says the traits that make someone an explorer, curiosity, adaptability, a willingness to relearn, don't track with rank or tenure.

The other 95%, in his framing, are exploiters, engineers who ship with proven methods rather than testing what AI can do past the official playbook. He doesn't offer a way to spot explorers in advance. His point is that you can't, so stop trying.

His fix is process, not talent-scouting. Let explorers self-identify by what they actually try. Pull their discoveries into documentation and training the rest of the team can use. Track how many engineers move from exploiter to explorer over time, instead of chasing a single 100x outlier.

The bet worth arguing with: if your AI story is still built around one star engineer's output, you're measuring the wrong 5%. Raghunathan's math says the win is moving the 95%, not celebrating the outliers.

He posted the argument on Stack Overflow's blog on August 5, aimed at leaders being pressed to prove AI gains with individual metrics.