Data shapes design
I’ve seen many teams go through the shift of shipping their first predictive or AI-powered feature.
One of the biggest mindset shifts required? Moving from “design shapes data” to “data shapes design.”
In traditional software, design shapes data. You design the UI and APIs you want to build, then create a schema to support them. Data conforms to the design.
It’s like building a house: you start with the blueprint, then pick materials to match the design. You might choose wood siding, a metal roof, and a concrete foundation. The design comes first, materials second.
In AI, it flips. Data becomes your raw material and your design constraint.
You’re still building a house, but whether you’re working with straw, wood, or steel completely changes what you can build. You can’t build a glass pavilion if you only have bricks.
Your data determines what’s feasible, stable, and valuable.
That’s why the first step in building any AI feature should always be a POC with your actual data:
- Define the problem you’re solving
- Check what data you actually have
- Validate if your data can support the experience you want
Once you validate it’s shippable, then design the surrounding system. Otherwise, you’re designing based on (very likely) false assumptions.