discovery — when you need to build to learn
Three tools for gaps that watching cannot close
Some unknowns cannot be resolved by observation. When the unknown is about whether an approach is technically feasible, or whether the direction chosen would make sense to the person observed — something needs to be built to answer the question. Three tools exist for this.
| Tool | When to use it | What it produces | What it does not produce |
|---|---|---|---|
| Spike | A technical unknown that cannot be resolved by reading documentation. "Can this approach handle our load?" Time-boxed: one to five days. | A written conclusion: what was learned, what the approach can and cannot do, what changes about the direction. Usually exits as an ADR or benchmark. | Production code. A Spike exiting with shipped code was an unplanned feature. Spikes running longer than a week have become research Epics. |
| POC | Technical feasibility of a proposed approach before committing to it. Build just enough to answer the feasibility question. | A written conclusion about what the approach can and cannot do. Noted in the brief's Decisions: "We ran a POC on this and found…" | A foundation for production work. POC code is exploratory and discarded. Building on top of POC code is a reliable source of technical debt. |
| Prototype | Product direction uncertainty before the brief is agreed on. "Would the direction we chose make sense to Gal?" Low fidelity. Shown to the witnessed person, not stakeholders. | A decision about the direction. Either confirmed or a misunderstanding revealed, requiring a return to What We Learned. The finding is noted in Decisions. | A design specification. A Prototype tests whether the direction is correct, not whether the design is final. |
How to apply this
- ✓ Before running any of these, state the specific question being answered. If it cannot be stated precisely, the tool will not produce a useful answer.
- ✓ All three exit with writing, not code or design assets. The written conclusion is the deliverable — it updates the brief's What We Learned or Decisions.
- ✓ All three are time-boxed. Name the time box before starting. If it ends without a conclusion, extend once with a stated reason.
- ✗ Do not run a Spike to avoid making a decision. If the question is already answered, another Spike is procrastination with technical packaging.