Discovery and Delivery Are Not the Same Sport

Discovery is divergent. Delivery is convergent.

In discovery, my job is to expand the surface area. I sit with fleet managers and listen to how they actually plan a route on a Monday morning. I run prototypes that I know are wrong, just to see which part of "wrong" the user laughs at. I write three competing problem statements and then throw two away.

In delivery, my job is the opposite. The decision is made. The design is locked. My only job is to remove ambiguity until the engineers can sleep through the sprint. Any new "interesting question" that arrives is a threat, not a gift.

A team that mixes the two ends up with engineers who feel scope-creeped to death and a PO who feels like nobody is curious enough.

What I learned shipping AI/ML driver mapping

When we built automatic driver-vehicle assignment in Reveal, the temptation was huge: "we have ML, let's just train and ship." We held the line on discovery for longer than felt comfortable — talking to dispatchers, looking at the messy reality of how vehicles get swapped at depots, watching how often the "wrong" auto-assignment would silently break their day.

That extra discovery is the only reason we hit ~90% accuracy in the prompts. Not the model. The framing of when the model should and should not speak.

The compliance trap

The other half of my work — SMOG, CARB, CPRA — is regulatory. Compliance features tempt you to skip discovery entirely because "the regulation already told us what to build." That's a trap.

The regulation tells you the *rule*. The user tells you the *workflow* the rule has to live inside. If you ship the rule and ignore the workflow, you'll get audited-clean software that nobody opens twice.

CARB certification was won in the lab. CARB *adoption* was won by sitting with the operations team and removing three clicks from the daily flow.

The rule I now use

When a stakeholder pushes me to "just decide and move," I ask one question: am I being asked to discover, or to deliver?

If the answer is delivery, fine — I'll move fast, I'll lock scope, I'll be boring.

If the answer is actually discovery dressed up as delivery, I push back. Politely. Once. And then again, because the second push is usually where the real conversation starts.

That single distinction has saved me more failed launches than any framework on a slide.