In the 1970s, Chris Argyris introduced the concept of double loop learning.
The idea is that when something goes wrong, you can either just fix the immediate problem (single loop learning), or go deeper and question the underlying assumptions and strategies that led to the problem in the first place (double loop learning).
It's a powerful framework for continuous improvement that has applications well beyond the realm of organizational learning.
Turns out, the best AI product teams often adopt a similar approach, whether they call it double loop learning or not.
That's because developing AI products effectively requires two distinct skill sets:
- Technical chops to build the thing, and
- Subject-matter expertise to make sure it actually does what it's supposed to do.
You've got your developers and ML engineers (technical) on one side, and your SMEs, PMs, and CX folks (experts) on the other. The magic happens when you create a tight feedback loop between the two.
But it's not as simple as just putting everyone in a room together and hoping for the best. You have to be intentional about how you leverage each skill set.
The technical side is all about efficiency - streamlining the AI pipeline, running evaluations and tests, and making sure the machine is humming.
The expertise side is more about steering the ship - prompting the right questions, manually reviewing outputs, and making sure the product is actually solving the problem it's meant to solve.
The key is to create a virtuous cycle between the two.
The technical folks build something, the subject-matter experts poke holes in it, and then the technical folks go back and make it better. It's a constant process of questioning assumptions, testing hypotheses, and iterating based on feedback.
Rinse and repeat.
In other words, it's double loop learning in action.
Of course, this is easier said than done. It requires a level of trust and collaboration that doesn't always come naturally, especially in cross-functional teams. But the payoff is worth it.
When you get it right, you end up with an AI product that not only works, but actually delivers value to the end user.
So if you're building an AI product, take a page from the double loop learning playbook.
Create that tight feedback loop between your technical and subject-matter experts. Question your assumptions early and often. And most importantly, keep iterating until you get it right.
It's not always a smooth process, but it's the best way to build something that truly makes a difference.