Discovery: Scoping Before Any Code
Discovery is the work of turning "we have a problem with X" into something specific enough to build. Skip it, and you build the wrong thing quickly — which is worse than building the right thing slowly.
A short discovery pass asks, in order
- What does success actually look like to the customer, in their own words, not yours.
- What data and access do you really have, versus what you were told you have.
- What is the smallest version of this that proves the idea works at all.
The trap discovery avoidsWithout discovery, teams often build the version of the problem that is easiest to code, not the version the customer actually has. The two can look similar in a planning meeting and turn out completely different once real usage starts.
Do this in 60 secondsTake this one-sentence problem: "our team loses track of customer requests." Write five discovery questions you would ask before touching any code — at least one about what success looks like, one about current tools, and one about who else is affected if you get it wrong.
Today’s AI angleBefore building an AI agent for a customer, discovery has to answer one extra question: what does the agent being wrong cost them, since that single answer decides how much human review the system needs.
Interview prepA stakeholder tells you "we need better reporting." How do you turn that into something buildable? I would not open an editor yet. I would ask what decision this reporting is meant to support, who looks at it and how often, and what "better" has failed to mean so far — usually there is a specific number someone could not get last time. That turns a vague request into one measurable thing to build first, instead of a reporting platform nobody asked for.