Cadence is freezing v1 scope on Friday. The open question: does AI-powered blocker-clustering go in v1, or does v1 ship manual tags and add clustering in v2? It reads like a feature call; it's really a positioning call with an engineering cost attached. We pinned it to one fork, one owner, one deadline — because a "decision" with no fork isn't a decision, and one with no date is a guess that never gets checked.
Ship manual tags in v1; instrument for AI clustering in v2. Manual tags deliver the core "see what's blocked" value on Friday's timeline; the AI layer is a differentiator, not the product, and it's cheaper and safer to add once real tag data exists to train and evaluate against.
Committed review date: the v2 planning cycle (≈8 weeks) — revisit with real usage data, or sooner if a competitor ships clustering first.
Confidence: MODERATE — the recommendation rests on one live assumption (below); the strongest case against it is real, and named.
| Option | The one-line case for it |
|---|---|
| A · AI clustering in v1 | Ships the differentiated "wow" on day one; the demo that drives early word-of-mouth. Highest upside, highest cost and risk. |
| B · Manual tags in v1, AI in v2 rec | Delivers the core value on the Friday freeze; defers the risky, data-hungry AI layer to when there's data to build it on. |
| C · Do nothing (no blocker view in v1) | The honest baseline. Protects the timeline completely, but ships a v1 that doesn't address the #1 user complaint. Weighed, then rejected. |
A fourth "buy a clustering API" option was generated and cut — it's really two decisions bundled (build-vs-buy is its own brief), and no vendor covers Cadence's blocker taxonomy.
| Criterion (weight) | A · AI in v1 | B · Manual, AI v2 | C · Do nothing |
|---|---|---|---|
| Hits Friday freeze (×3) | At risk | Yes | Yes |
| Solves the #1 user need (×3) | Fully | Core of it | No |
| Differentiation / "wow" (×2) | High | Deferred | None |
| Build risk / eval burden (×2) | High | Low | None |
| Weighted read | Strong-but-fragile | Best balance | Safe-but-empty |
The matrix points at B — but a single number never decides a nuanced call. It informs the recommendation; the judgment (and the case against) makes it.
The strongest honest argument against B: shipping manual tags makes Cadence "the manual one." The day-one AI "wow" is precisely what earns the early demos, press, and word-of-mouth that manual tagging never sparks — and that flywheel is hard to restart in v2. If clustering is cheaper to build than we think, and a competitor leads with it, we will have traded the narrative to save three sprints.
A confident AI answer would have handed you option B and stopped. The case against is the deliverable — it's what lets you defend the call later, or change it early.
AI expands the option set and drafts the steelman fast. It cannot rate its own honesty — so the discipline is forcing the moves a satisfying answer skips.
| Ask an AI "what should we do?" | This brief |
|---|---|
| One confident answer, instantly | Real options, including do-nothing |
| Assumptions smuggled in as facts | Known / assumed / unknown, split honestly |
| No sense of how sure it is | A stated confidence level |
| Never argues against its own pick | The Strongest Case Against, mandatory |
| Feels like a decision | Is one you can defend — or revisit on a date |
A recommendation you can't defend later isn't worth paying for. The structure is the product.
One decision, framed the way a $500/hour consultant would — real options, honest trade-offs, a confidence-rated recommendation, and the case against it. Delivered within 2 business days. Book a Decision Brief →