This retires the buyer maps in issue #4 and issue #5. Same five published cases. Same core archetypes. The sorting axis is new: altitude, the height from which a given evaluator inspects.
This is a directional call, not a measured one, and I show my calibration two sections down. Agent work in senior portfolios is commoditizing at the level of it exists. Not at the level of it works. Six months ago, having AI casework was the separator. It has stopped being one. The separator now is whether the evaluator can see the mechanism under the screen, and which evaluator is in the chair decides which mechanism you put in front of them.
The 30-second matrix
| Buyer in the room | Altitude | Lead | Support | Cut / Hold |
|---|---|---|---|---|
| Founder-CEO (seed–Series B) | Craft | Carrier IQ correction flow. The quote comes back wrong; a human takes it back | Equinox+ (0→MVP, 3 months, 5 brands) | Cut: Alibaba org narrative. Hold: Thermo Fisher governance |
| CPO / VP Product (hiring manager) | System | Alibaba: the mandate you built before anyone assigned it | Thermo Fisher mySupply ($20M+ margin, 12 months) | Cut: Trust essay conceptual sections. Hold: exactly one handoff |
| VP / Head of Design (peer or manager) | Leadership | Five handoffs as an enforced standard, shown inside a Labs implementation | Thermo Fisher (100% adoption, 6 pharma partners) | Cut: Lab demos beyond one. Hold: Alibaba GMV |
| Design panel (senior ICs, veto power) | Craft | One Agentic Labs component, every state, full iteration history | Allē dual-surface (30M members, 3.2× redemption, $42 CAC) | Cut: business outcomes. Hold: team structure |
| Clinical AI evaluator (physician-founder, informatics lead) | System | Thermo Fisher mySupply (regulated, 6 partners) | Red Cross ($847K disbursed, 6 systems→1, national) | Cut: Alibaba. Hold: Carrier IQ correction flow |
| Cross-functional panel (eng/PM leads) | System | Thermo Fisher architecture | Red Cross, craft deep dives | Cut: business framing. Hold: outcomes |
The cross-functional panel card survives the altitude reframe intact. System altitude, Thermo Fisher leads, architecture ahead of outcomes. It lives in issue #4. I build a new card only when the angle actually moves.
Calibration, and one retraction
High confidence. Six-month decay. Designer Fund's 2026 survey of 900-plus designers puts weekly AI use in design work at 91%, up from 54%. Half have shipped AI-generated code to production.
Moderate confidence. Figma's hiring research has 79% of hiring managers reporting increased need for AI product design knowledge. That measures demand. It tells you nothing about supply.
Low confidence, and this is the hole in the argument. Nobody publishes a count of how many Director+ or Staff candidates now show agent-interface or trust casework. One practitioner describes senior designers moving AI work to the front of their portfolios across the first half of 2026. Posted in July, so treat it as current priority rather than validated pattern. No sample size. Directional at best.
Now the retraction. In issue #4 I told you most candidates at your level arrive with a strong portfolio and no framework. That was positioning language wearing the costume of a finding. I had no measurement behind it then and I have none now. Drop the line. Use the Trust essay because it holds your argument together under pressure, not because I told you the field is missing one.
Craft, system, leadership: what each height inspects
Craft. One surface, viewed close.
They want the decision underneath the pixel. Shopify's design recruiter, asked what is hardest to find, named discernment, and marked a candidate down for shipping a generated result unelevated. Figma's VP of product design says the fidelity bar jumped the moment prototypes got cheap.
- Your moment: one Agentic Labs component across every state, plus the versions you killed and what killed them.
- Supporting case: Allē, where the dual-surface constraint forced craft calls against two audiences at once.
- Competitors bring: clean AI-assisted output with no visible authorship.
Match by showing the discard pile.
System. Across surfaces, viewed for coherence.
A product design recruiter reports that Staff and Principal portfolios fail most often on isolated features with nothing proving the work held at scale. Some clients now ask for raw files and clickable prototypes rather than a narrated deck.
- Your moment: Alibaba as three sprints running on one architecture across homepage, search, and PDP. Not three projects that happened to share an employer.
- Supporting case: Red Cross. Six systems into one. Coherence as a literal deliverable.
- Competitors bring: three good features.
Exceed by showing the primitive that generated all three.
Leadership. Viewed for whether the standard outlives you.
The same recruiter tracks "player-coach" and "leading through the work" spreading across director searches, with recent hands-on involvement treated as non-negotiable.
- Your moment: the five handoffs as a written standard.
- Supporting case: Thermo Fisher. Six pharma partners at 100% adoption against it.
- Competitors bring: headcount growth.
Exceed by showing a judgment somebody else applied with you out of the room. That framing is mine, not the market's. The sources say quality uplift and platform thinking. Take my idea into the room in their vocabulary.
Play cards by buyer
Founder-CEO
Who this is. Seed through Series B. Frequently the only real decision-maker. Shows up first and last, and will run over the process if they like you.
What they're evaluating. Craft and leadership compressed into a single forty-minute read. The discernment gap the Shopify recruiter describes, plus one question: can you ship without supervision.
- Lead with Carrier IQ. Specifically the correction flow. Not the architecture. The recovery.
- Support with Equinox+. 0→MVP in three months across five brands answers "can you compress" with a number rather than an adjective.
- Cut the Alibaba org narrative outright. At this stage it reads as bureaucracy. Hold Thermo Fisher governance for the moment enterprise buyers come up.
- IC+manager tilt: 80/20 IC. Open a live Lab. Do not open a slide.
Landmine. "Head of Product means you stopped designing." It lands in the first ten minutes. Reframe: you went to product to build AI-natively from zero, you are in agent traces daily, and you are coming back to design for vertical depth over horizontal platform. TinyFish is currency in this room. Never collateral.
Competitive field. Design engineers who will write code in front of the founder. You will not out-code them. Out-judge them.
CPO / VP Product
Who this is. Owns the requisition. Sets your level. Usually the hardest hour in the loop.
What they're evaluating. System altitude, measured against the isolated-features failure above. Can you carry a mandate that nobody has written down.
- Lead with Alibaba as mandate creation. You named a structural gap with no owner, built the mandate, ran three cross-functional sprints. Headline is +20% daily transactions and +2.2pt NPS. The $50B+ GMV is context, not the point.
- Support with Thermo Fisher mySupply. $20M+ margin, six pharma partners, twelve months. It proves the pattern repeats inside regulation instead of depending on one company's culture.
- Cut the Trust essay's conceptual sections. A product hiring manager reads framework exposition as a candidate who has thought more than she has shipped. Hold exactly one handoff for the moment they ask how you would approach their AI surface.
- IC+manager tilt: 50/50. Show a design decision that changed the roadmap sequence. Not the team topology slide.
Landmine. "Builder, not a strategist" or the inverse, depending on how the recruiter set you up. It surfaces after your third execution story. Reframe by naming the gap before you describe what you shipped into it.
Competitive field. Candidates with bigger org charts. Beat them on whether design changed what got built, not on how many people built it.
VP / Head of Design
Who this is. Your peer or your manager. The most technically skeptical reader in the process, and the one most likely to catch narration standing in for work.
What they're evaluating. Leadership altitude, hands-on quality uplift specifically. Does hiring you raise the craft bar or just add capacity.
- Lead with the five handoffs as an operating standard, shown where it is enforced rather than where it is described. Use the published Agentic Labs implementation, Carrier IQ's correction sequence, because they can open it themselves. Then say out loud that you run the same handoff against Fortune 500 deployments today. Do not put it in the deck. TinyFish is not published work, and a design leader will ask to see the case study you cannot show.
- Support with Thermo Fisher. 100% adoption across six partners is a standard other teams complied with, which is the closest your published work comes to proving your judgment travels.
- Cut the Lab demos down to one. This buyer hears a demo reel as sales. Hold Alibaba's GMV until asked. Scale numbers land as deflection with design leaders.
- IC+manager tilt: 60/40 manager, with a craft artifact under every management claim. Show a design review you ran that changed a decision, and show the before.
Landmine. "Player-coach on paper, coach in practice." It comes in round three, when every hands-on example turns out to be eighteen months old. Bring one thing you personally designed in the last ninety days.
Competitive field. Design leaders with visible team-quality uplift. Match them by naming the standard, then naming who else applied it.
Design panel (senior ICs)
Who this is. Two to four senior or staff designers. No hiring authority. Complete veto power.
What they're evaluating. Craft altitude, and nothing else, per the discernment gap above. Are your hands current.
- Lead with one Agentic Labs component in all states, iteration history included. Brand Pulse or Retail Velocity, whichever has the cleaner set of rejected versions.
- Support with Allē. 30M members, 3.2× redemption, $42 CAC. Proof the craft held while serving two audiences simultaneously.
- Cut every business outcome from the main flow. Hold them for the one panelist who asks.
- IC+manager tilt: 90/10 IC. Bring the file, not the deck.
Landmine. "She's been managing too long to have current hands." It arrives disguised as a question about a single interaction detail. Pre-empt it. Open on a pixel decision, not a strategy frame.
Competitive field. Agent screens are no longer rare. That I will assert. Whether the state machine underneath and the discarded versions are also rare, I cannot measure, because no public data counts what senior candidates actually publish. Call it speculative, then act on it anyway. Showing the mechanism costs you nothing if the field already shows it, and wins the card if it doesn't. Note that this is a different reason than the one I gave in issue #4 for the identical recommendation.
Clinical AI evaluator
Who this is. A physician-founder, a clinical informatics lead, or a design leader with a clinician sitting next to them. Appears mid-process, and the schedule rarely tells you they're coming.
What they're evaluating. System altitude with a domain gate bolted on top. Do you understand that a wrong output has a licensed human's name attached to it.
- Lead with Thermo Fisher mySupply. Regulated pharma, six partners, twelve months, 100% adoption. Maven's posting accepts "regulated domains" alongside health tech. Layer Health accepts "analogous experience." That language is your textual permission to lead with a non-clinical case.
- Support with Red Cross. $847K disbursed, six systems into one, national in six months. Frame it as eligibility, authority, and auditability under time pressure. Do not frame it as healthcare.
- Cut Alibaba entirely. B2B marketplace scale invites the read that you optimize for throughput while they optimize for not being wrong. Hold the Carrier IQ correction flow for the moment they ask how a clinician overrides the model, which they will.
- IC+manager tilt: 70/30 IC. Walk the record. Who saw what, who signed, what the audit trail shows afterward.
Landmine. "She has never designed for a clinician." It comes from the physician, and it is fair. Adjacent reads as credible; it does not read as equivalent, and claiming otherwise costs you the room. Say instead that you have designed the moment a licensed professional accepts responsibility for a system's output, and that month one goes to shadowing clinicians.
Competitive field. Candidates with real EHR or clinical documentation history. You will not beat them on domain. Beat them on the mechanism, which most of them have designed exactly once, for one workflow, inside one system.
Moderate confidence on the domain gate. Across six current clinical AI postings I checked (Abridge, Ambience, Spring Health, Maven, Layer Health, Omada), healthcare background sits in bonus or preferred language, not in requirements. Every one of them still expects rapid immersion in clinical workflow alongside clinical experts. Adjacency clears the application. It does not settle the room.
When the room is a hybrid
Most loops mix these buyers. The founder-CEO card and the design panel card point in opposite directions.
Resolve it by the person who can kill you, not the person who can hire you. Panels veto and founders override, so give the veto-holder their altitude and give the decision-maker your bridge.
If you cannot read the room in advance, ask the recruiter who sits in the portfolio session and what they usually push on. That answer names the altitude. Vague answer, default to system and open with Alibaba. It plays acceptably at all three heights, which is precisely why it should never be your first choice once you know who is actually across the table.
- Autonomy and interruption move together: Anthropic's measurement of agent autonomy found experienced users auto-approving more (20% → 40%+) while interrupting more often, which means the Watch → Verify → Delegate ladder in your Trust essay is hiding a two-axis behavior — autonomy scope and intervention sensitivity are separate controls, and the visual you build should show both.
- Altana's posting is the vocabulary source: Its Head of Product Design listing describes the design system as a machine-readable language of patterns and interaction grammars that PMs, engineers, and agents compose — the clearest statement yet of what a system-altitude evaluator will ask you to prove, though the company's last confirmed round is a 2022 Series B, so the equity window is unverified.
- Evals as a design deliverable: First Round's account of Figma's AI evaluation process documents one-to-four design and functionality scoring, a ~1,000-example corpus, golden prompts, and nightly model comparisons — this is the vocabulary behind "what does done mean for a probabilistic product," and your LangSmith-to-design-change loop is the answer you cannot yet show publicly.
- Fix the gate before the next artifact ships: Your private essay page still delivers full HTML to unauthenticated browsers before the JavaScript password check runs, which means an evaluator inspecting your portfolio's own trust boundary finds an interface that says restricted and a system that says delivered — repair it before publishing anything new about trust design.