Working with AI

The arrival of AI has dramatically accelerated development. At the same time, when used incorrectly, it can mass-produce low-quality systems at speed — and in the worst case, severely damage a business.

AI is not magic. It is a convenient tool, like a knife or a car. A knife alone cannot create a delicious dish without a recipe. A car alone cannot reach a destination without one. In the same way, AI alone cannot build valuable services. The same tool produces vastly different results depending on the hands that wield it.

We believe what matters is that people who understand AI's strengths and weaknesses assign it the right role in the right place. We separate the work as follows:

• Upstream work (requirements, design, judgment) is led by experienced humans. The quality of the specifications produced here sets the ceiling for what AI can deliver.
• Implementation and testing are driven by AI — fast iteration at scale. Humans review the output and judge whether it aligns with design intent.
• Building prototypes early, we let users interact with the actual product repeatedly. We move toward the right answer through dialogue with real artifacts, treating scrap-and-build as the default mode.

For AI to produce high-quality outcomes, it requires equally high-quality design and judgment, layer upon layer. The value we offer is not in commissioning AI — but in being the people who know how to draw value out of it.