The Craft Judge Room

The craft judge is in the room to answer one question: does she still do the work?

You have evidence for every version of an affirmative answer — IC work, team leadership, enterprise case studies at BCG Digital Ventures, forward-looking artifacts for frontier AI problems, a shipped production AI product. The evidence isn't the constraint. "Hands-on" means something different to a senior IC at an AI-native company than it does to a design director at an enterprise platform, and answering the wrong version spends the only five minutes where the room's attention is uncommitted.

The Evaluator Map covers this room at survey level: the feared failure is hiring someone whose hands-on judgment decayed while she was managing. Take that as given. What follows is the positioning complexity your specific background creates inside it — a hybrid IC-manager record with a recent Head of Product title sitting on top of a design career, and AI production work that isn't in the published portfolio.

Tension 1: What "hands-on" means to the person across the table

The word appears in nearly every senior design posting, and it means something different in each.

AI-native companies. The craft judge is usually a senior IC or a design lead who builds things most days. Hands-on means builder proximity: you prototype, you make interaction decisions at the level of behavior, you can reason about what's technically cheap and what isn't. OpenAI's Codex posting asks for recent production-code shipping, high-fidelity prototypes, and comfort working next to engineers and researchers. This person is not assessing whether you can judge someone else's design. They want to know whether you would make the design. High confidence — the pattern holds across every AI-native posting I've tracked.

At AI-native companies the craft judge may also not occupy a session of their own. OpenAI's interview guide describes a four-to-six-hour final loop and never names a dedicated portfolio or craft review. Craft evaluation is probably distributed across the whole loop, which makes every interviewer partly a craft judge. Moderate confidence — I'm inferring from an absence in published material plus the general pattern that smaller, engineering-heavy companies assess design across conversations rather than isolating it in one.

Enterprise platforms. The craft judge is more often a design director or senior manager, and what they're measuring is craft calibration: whether you can set a quality bar across a product surface and hold it through other designers' hands. Atlassian runs the same 45-minute portfolio review for IC and management candidates, then splits the follow-ups. ICs get pressed on the design choices they made. Managers get pressed on how they led and shaped the result. The design still has to be on the screen. High confidence; Atlassian publishes the distinction and CZI's structure reinforces it.

Growth-stage companies. Hands-on usually means something blunter: you are not above the work. Ramp's Director posting wants the leader to begin by owning a product area and shipping alongside designers, PMs, and engineers. Headway's Staff posting expects code prototypes and involvement through launch. This craft judge screens for willingness as much as ability. Moderate confidence, inferred from posting language rather than published criteria — though language like that usually follows a senior hire who arrived and immediately delegated.

The taste-first variant. Suno's Head of Product Design posting, now closed but preserved here, set its bar with "simple, joyful, and intuitive," "exceptional taste," and "emotional resonance." That's a third axis, separate from builder proximity and craft calibration. When a posting leans on joy, delight, polish, or taste more than on systems thinking or technical fluency, expect a craft judge evaluating aesthetic conviction first. Moderate confidence: Suno is the cleanest example in current data, and I've seen adjacent language at other consumer-creative AI companies but not enough of it to call this a stable category.

The cross-functional assessor. Sometimes the person judging your craft isn't a designer. An engineering lead, a PM, or a researcher may sit in the craft session or carry a craft lens into their own. Each filters "hands-on" through their discipline: the engineer wants your judgment to account for implementation cost, the PM wants design decisions tied to user outcomes and business goals, the researcher wants evidence rather than assumption. You'll recognize them by where the questions land — trade-offs, feasibility, measurement, rather than interaction and visual choices. Render's Agent Experience posting covers CLI, API, SDK, MCP, and dashboard design, a scope that all but guarantees engineering representation in the evaluation. Moderate confidence on the pattern. High confidence on the instruction: ask the recruiter who is in the room.

Reading the room in the first two questions

• "Walk me through a design decision you made recently." Builder proximity. AI-native or growth-stage.
• "How did you raise the craft bar on your team?" Craft calibration. Enterprise.
• "Show me something you're proud of." Unresolved. Wait for the follow-up before you commit to an altitude.
• "What's your design philosophy?" Taste. Lead with conviction about quality, not with process.
• "How would you approach this differently given [technical constraint / timeline / data]?" Cross-functional. They're testing whether your judgment holds up inside their discipline's reality.

If the signal doesn't arrive, ask for it: "Would it be more useful to walk through a design decision I made hands-on, or how I shaped a team's output?" That reads as fluency rather than evasion, and it gets you the version they actually want.

Tension 2: Which evidence earns credibility here

Three layers matter in this room and they are not interchangeable. Your published BCG Digital Ventures case studies are enterprise-scale, shipped, and inspectable — a reviewer can open them without you. Your forward-looking artifacts for frontier AI problems are current and calibrated to the problems AI-native teams are hiring against, but they aren't linked from your portfolio's main navigation and can't be examined with you absent. TinyFish is production-AI context that establishes technical credibility and never functions as portfolio proof. Prior coverage maps what each of these can and can't carry.

Lead with production cases when the judge is measuring calibration or willingness. Enterprise craft judges and most growth-stage craft judges want to see what happened when your design decisions met a real constraint set: the trade-offs, the measurable outcome, the evidence that the work shipped and mattered at scale. The BCG Digital Ventures cases carry that. Lead with them when the posting emphasizes measurable impact, platform coherence, or scaling design across a surface.

Lead with forward artifacts when the judge is measuring builder proximity on an unsolved problem. Trust architecture, agent coordination, model-behavior transparency — an AI-native craft judge working on any of these usually cares more about how you think about the unsolved case than how you executed a solved one. Lead here when the posting emphasizes 0-to-1 work, novel interaction patterns, or AI-native product thinking. One constraint to plan around: these artifacts only work live. Don't reference them in written outreach and don't expect anyone to find them on their own.

For the taste-first judge, neither layer leads on its own terms. The BCG Digital Ventures work shows enterprise rigor, not emotional resonance. The forward artifacts show frontier thinking, not taste. Pick whichever single project, shipped or speculative, best shows an aesthetic conviction you held and defended, and lead with the conviction rather than the complexity. If nothing in your current portfolio carries that, treat it as a fit signal and take it seriously.

For the cross-functional assessor, lead with the project where the trade-off was most visible and most expensive. Show the decision, then show what you gave up to make it and why that exchange was right. The BCG Digital Ventures work is strong here because enterprise design is mostly constraint negotiation. This person doesn't need to admire the craft; they need reasoning that survives pressure from their side of the table.

Order matters as much as selection. Evidence order shapes the interrogation path. Open with a production case in an AI-native room and the follow-ups go to enterprise decisions that never touch the judge's real concern, spending your freshest window on work that doesn't differentiate you. Open with a forward artifact in an enterprise room and the judge anchors on the fact that it hasn't shipped, then uses the follow-ups to test whether you can survive production. Moderate confidence — this comes from cross-company evaluation structures and candidate outcomes, not published scoring rules.

TinyFish is the bridge between the enterprise production record and the AI-native positioning. When a craft judge asks how you've worked with AI systems in production, TinyFish is the answer, delivered as context in ninety seconds rather than as a walkthrough. It settles whether you have actually built AI things. It does not substitute for showing design.

Sequences

• AI-native. Open with a forward artifact demonstrating interaction judgment on a frontier problem. When the judge probes, use TinyFish to ground the thinking in production. If they want shipped work, pivot to the BCG Digital Ventures case whose constraint profile most resembles their product's.
• Enterprise. Open with the BCG Digital Ventures case showing the most complex judgment under real constraints, metrics included — adoption, error reduction, workflow completion, whatever you measured. When AI or current practice comes up, introduce TinyFish as context and a forward artifact as evidence of where your thinking is now.
• Growth-stage. Open with shipped work and emphasize velocity and personal involvement over organizational scale. This judge wants to see that you built the thing, not that you directed the building of it.

Tension 3: The altitude trap

This one is hardest to prepare for because you solve it live.

Narrate at a strategic altitude — built the org, set the quality vision, directed the team toward the outcome — and the judge starts wondering whether you touched the work. Narrate at a tactical altitude — this type scale, this error state, this flow iteration — and a leadership-evaluating judge starts wondering whether you can operate above the details. A candidate who is cleanly one thing has an easier time here. You've done both, so the instinct is to show both at once, and work presented at two altitudes simultaneously usually registers at neither.

What published structures reveal

CZI's interview guide makes the mechanism visible. For each project you name your collaborators, state what you did versus what others did, show the final interaction and visual design, explain the trade-offs, and separately demonstrate leadership and organizational impact. Design Excellence and Leadership are scored as distinct categories. Strength in one does not compensate for silence in the other.

Atlassian arrives at the same separation differently: one portfolio review, follow-ups that shift by track, and a management candidate who still has to show the design result.

High confidence that these two represent a real pattern in how companies separate craft evaluation from leadership evaluation. Lower confidence on how widespread it is — most companies don't publish enough to check.

Drift signals and recovery

Strategic drift:

• The judge asks you to zoom into a specific design decision and you answer with the team's process.
• You've said "we" three times without saying "I."
• The questions keep getting more granular. That's a test of whether you know the work.

Recovery: stop narrating the organization. Take one decision — the most consequential interaction choice, the hardest visual trade-off — and reason through it at the level of the artifact. Show the thing. Say why this version and not the other one.

Tactical tunnel:

• Attention drifts during your third minute of interaction rationale.
• They ask what the impact was and you don't have it ready.
• The questions keep getting more abstract. That's a test of whether you can see above the details.

Recovery: connect the decision you just showed to the outcome it produced or the constraint it resolved, in one sentence. Then widen — what the decision taught you about the product's design direction, and what the team did with that next.

The taste-first version of this trap runs on a different axis: process-heavy versus conviction-forward. Spend your time on methodology, how you researched and tested and iterated, and this judge disengages. Recovery is to stop explaining how you got there and say what you think is right about the result. Speculative; I'm reasoning from the Suno posting's language, not from observed evaluation behavior.

The cross-functional version: an engineering lead or PM doesn't want visual rationale at designer depth. When their attention goes, pivot to the exchange — what the decision cost in engineering effort, timeline, or scope, and why it was worth paying. Moderate confidence.

Using the hybrid properly

A pure IC can't show leadership impact credibly. A pure manager often can't show hands-on judgment credibly. You can show both, provided you sequence them instead of blending them. Present one project at craft altitude, your own decisions specific and visible. Present a second at leadership altitude, the organizational impact clear. Each project carries one burden.

CZI asks for two end-to-end projects in 35 minutes, which is exactly the shape this needs. Use it even where the company doesn't prescribe it.

Before you walk in

Ask the recruiter who is in the craft session. Most companies don't publish their loop structure, which leaves the recruiter as your only source on who evaluates craft and in what format.

Read the posting for the construct, not the company name. A posting built around model behavior, evaluation infrastructure, and data strategy screens for different craft than one built around interaction design, prototyping, and visual quality — at the same company, in the same quarter. Match your evidence to the posting, not to the logo. Prior coverage maps which evidence answers which feared failure across tiers.

If the posting spans CLI, API, SDK, MCP, and dashboard design — as Render's Agent Experience role does — the judge is evaluating cross-interface coherence: one journey held together across surface boundaries. Your enterprise cases with cross-product or cross-platform decisions are the strongest thing you have for it.

If you can't identify the judge beforehand, default to the CZI sequence: state your personal contribution, show the artifact, explain the trade-off, address leadership impact separately. It isn't optimized for any particular version of this room, and it doesn't lose the room in either direction.