Systematic Discovery

When we encounter a new problem or opportunity, our reflex is to select and commit to a solution as quickly as possible. We might spend a few days brainstorming, fill a wall with sticky notes, interview some customers, and compare a small handful of alternatives. Then we build. And like clockwork, the solution in which we felt so confident fails - not necessarily because we are untalented, nor because we implemented it poorly, but because the failure began one step earlier: we treated a search problem as an execution problem.

For any given problem or opportunity, we cannot know a priori whether a viable solution exists, much less what shape it takes. What exists instead is a vast space of possible answers: some obvious, some improbable, most useless, and, if we’re lucky, a small subset capable of creating substantial value. The purpose of discovery is to explore enough of that space to learn which possibilities deserve commitment. Yet most of our methods drive us toward the first one that feels correct. We treat confidence as evidence, commitment as progress, and ignore the alternatives we never explored as inconsequential.

Discovery should be treated as a discipline in its own right. If development can become more systematic, measurable, and effective, discovery can too. We can become better at generating meaningful possibilities, preserving useful differences between them, testing what matters, learning from failure, and converging only when the evidence warrants it.

My research follows two connected paths:

• Organizational discovery asks how institutions can be structured to discover well: how they allocate capital, compose teams, protect divergent work, and decide which possibilities survive long enough to be tested.
• Computational discovery asks how new tools and methods can help us construct, search, evaluate, and evolve far larger spaces of possibility than unaided intuition could manage.

The cost to translate an idea into a working prototype is rapidly approaching zero, but expanding the range of possibilities is not the same as discovering which possibilities matter, what evidence to trust, or when to converge. AI does not solve discovery. It makes systematic discovery possible at a scale we could not previously afford.