I spent a year as the bridge between an ML research team and a product engineering team.

It was the most challenging and educational experience of my career.

The researchers: brilliant, creative, published at top conferences. Their code worked in notebooks. Their experiments were fascinating. Their prototypes couldn't handle 100 concurrent users.

The engineers: pragmatic, reliable, shipped production systems. They thought ML was a black box. They wanted clear APIs and SLAs. They didn't understand why the model "sometimes" gave different answers.

The communication gap was vast. Researchers would say "the model achieves SOTA on the benchmark." Engineers would ask "what's the p99 latency?" Both would look at each other confused.

What I learned about bridging this gap: create explicit interfaces. The research team produces a model artifact with documented inputs, outputs, and performance characteristics. The engineering team consumes it through a standardized API. Neither needs to fully understand the other's work.

Establish shared metrics. Not academic metrics AND engineering metrics — shared ones that both teams care about.

And respect the different cultures. Researchers optimize for novelty and understanding. Engineers optimize for reliability and maintainability. Both are necessary. Neither is more important.

The organizations that build the best AI products are the ones that bridge this gap well. It's a rare and valuable skill.