# The Punctuated Discovery Sprint Template

**A 3-week intense discovery sprint sandwiched between 8-week build phases.**

By Falk Gottlob — falkster.com

---

## When to use this template

For AI-native products where continuous discovery (1-2 hours/week of customer interviews, every week, forever) is too slow. For traditional SaaS, continuous discovery still works — see [/handbook/continuous-discovery-autopilot](https://falkster.com/handbook/continuous-discovery-autopilot).

The cycle: 8 weeks build → 3 weeks discovery → 8 weeks build → 3 weeks discovery → etc.

---

## The 3-week sprint structure

### Week 1: Customer interviews (40% of allocation)

| Day | Activity |
|---|---|
| Mon | Sprint kickoff. Confirm interview targets. Calendar booked. |
| Tue-Thu | Customer interviews. 8-12 customers. Each 45-60 minutes. |
| Fri | Interview synthesis. Pattern detection. Open questions logged. |

**Targets:** mix of customers (heavy users, light users, churned, prospects who didn't buy). Goal: surface jobs-to-be-done, dispute patterns, willingness-to-pay anchors.

### Week 2: Agent testing (40% of allocation)

| Day | Activity |
|---|---|
| Mon | Define test workloads. Pull representative customer inputs. |
| Tue-Wed | Run agent against test workloads. Watch where it drifts. Build evals on the gaps. |
| Thu | Stress test with adversarial inputs. Compare behavior across model versions. |
| Fri | Synthesis. Surface the agent surface that customer interviews missed. |

**Targets:** at least 100 real customer inputs run through the agent under test. At least 5 new evals shipped.

### Week 3: Synthesis and competitive surface (20% allocation each)

| Day | Activity |
|---|---|
| Mon | Competitive scan. What did competitors release in the last 90 days? What did the underlying foundation models add? |
| Tue | Ecosystem scan. What did customers' other AI tools do that changed expectations? |
| Wed | Cross-source synthesis. Reconcile customer interviews + agent testing + competitive findings. |
| Thu | Draft the commit document. |
| Fri | Commit document review with team. Lock the next 8-week build commitments. |

---

## The commit document template

The artifact that ends every discovery sprint.

```markdown
# Discovery Sprint [Date]: Commit Document

## What we learned (5-10 findings)

For each finding:
- One-paragraph statement of the finding
- Source: customer interview / agent testing / competitive / ecosystem
- Confidence: high / medium / low

## What we'll test (3-5 hypotheses)

For each hypothesis:
- The hypothesis statement
- The eval or experiment that would validate or falsify it
- The owner
- The decision deadline

## What we'll ship (2-3 commitments)

For each commitment:
- The commitment in one sentence
- Success criteria (specific, measurable)
- The team owner
- The ship deadline (within the next 8 weeks)
```

---

## What gets cut from the regular calendar

During the 3-week discovery sprint:
- No feature work (engineering pauses on new features, focuses on shipping evals).
- No PRDs (the commit document is the artifact, not a separate PRD).
- No customer-facing meetings except discovery interviews.
- No new prototyping (use existing prototypes for testing).

If something on the calendar can't be cut, you're not running a real discovery sprint.

---

## Common failures

**One: customer interviews dominate.** The team falls back to 90% interviews, 10% agent testing, because interviews are familiar. The cycle becomes a more-intense version of continuous discovery without the agent-testing benefit. Fix: pre-block 40% of the sprint for agent testing on the calendar before the sprint starts.

**Two: no commit document.** The sprint produces "interesting conversations" but no decisions. The build phase starts ambiguous. Fix: the commit document review on Friday of week 3 is non-negotiable.

**Three: discovery sprints get skipped when "we're behind on shipping."** The team rationalizes that they need the build time. Six months in, the product has lost touch with the customer. Fix: the discovery sprint cadence is sacred. If shipping is behind, scope the build phase smaller — don't skip discovery.

---

*This template pairs with [Continuous Discovery Doesn't Scale for AI-Native Products](https://falkster.com/blog/punctuated-discovery-replaces-continuous).*