Start with the workflow, not the model

Teams often start AI products by picking a model and asking what it can do. That approach produces impressive demos and fragile products. Durable AI work starts with the workflow.

Before writing code, I map the user's goal, the friction they hit today, and the moment where intelligence could remove steps — not just add commentary.

Intent before intelligence

A workflow-first mindset forces you to define success in user terms: time saved, errors reduced, decisions made faster. Those metrics shape what the model actually needs to do.

This is especially important in operational tools like dashboards and assistant experiences, where users care about reliability more than novelty.

Design for adoption

If an AI feature requires users to change how they already work, adoption will stall. The strongest products embed intelligence into existing habits.

Start with one high-value moment in the workflow, ship it end-to-end, and expand only after users trust the first loop.