The first Counter-Narratives dossier on AI-native credibility named the grain of truth: Juno has not shipped from inside a frontier lab, and Agentic Labs is not equivalent to building Claude-scale interaction systems inside Anthropic. That piece recommended answering the objection with the full evidence stack: Agentic Labs, "Trust Is the New Interface," the agentic codas in Alibaba and Thermo Fisher. Correct recommendation. Incomplete framing. It treated "lab-native" as one objection when it's actually two different hiring problems using the same vocabulary.
One of them is a genuine mismatch. The other is Juno's strongest positioning. Know which is which before you write the outreach.
The Two Problems
Problem one: interaction-pattern invention for ambiguous model behavior. The model is the material. The designer sits alongside researchers. The output is paradigms, not interfaces. The question: "What should this AI's behavior even be, and how do we shape it through design decisions that are indistinguishable from research decisions?"
Problem two: trust, delegation, and control architecture for AI systems entering consequential workflows. The company has a model or uses one. The designer sits alongside product and engineering. The output is systems that make delegation safe. The question: "When does a human intervene, what do they see when they do, and what happens when they override?"
Both postings will say "AI-native" and "first principles." Both will mention "new interaction paradigms." The shared language doesn't sort them. The differences in everything else do.
Five Sorting Signals
Three of five to categorize with confidence. Run these against every AI design posting that hits your radar.
1. Where does "research" appear? In the team name or role title → lab-native. In the responsibilities as "use research to inform decisions" → applied. Being embedded in research and consuming research are different jobs.
2. What is the design material? Model behavior itself (shaping what the AI does, how it responds, what its defaults are) → lab-native. Model output in a workflow (what users see, how they evaluate it, when they intervene) → applied trust.
3. What verbs dominate? Explore, study, prototype, publish, define → lab-native. Ship, scale, lead, partner, drive → applied. I've written before about "built" vs. "designed" as a verb-level screen. Add "research" vs. "ship" to the same diagnostic layer.
4. What's the success metric? Capability growth, understanding, paradigm contribution → lab-native. Adoption, quality bar, speed, user trust → applied. A posting that says "measured by capability growth rather than engagement metrics" is explicitly opting out of product metrics. Take that to the bank.
5. What's absent? No team management, no cross-functional leadership, no design-system scale → likely lab-native IC embedded in research. No published-work requirement, no research methodology, no cognitive science → likely applied product design that needs AI fluency, not AI research capability.
Two Specimen Texts, Sorted
Anthropic's Design Engineer, Education Labs is the clearest lab-native posting I've tracked this year. Three signals land in the first two paragraphs.
The role description:
"part researcher, part product builder, part interaction designer"
Researcher comes first. That ordering is hierarchical, not alphabetical. The posting draws an explicit line between exploring new interaction patterns and optimizing existing ones, then tells you which side it wants. Success metric: "capability growth rather than engagement metrics." Preferred qualifications include learning sciences, cognitive science, HCI, educational psychology, published work on human-AI interaction or product philosophy. A research role that ships, not a product role that researches.
Signals 1, 3, 4, and 5 all point lab-native. Clean sort.
OpenAI's Product Design Manager sorts the opposite direction. (Caveat: this posting did not appear on OpenAI's current careers page as of this writing. May be filled, paused, or restructured. The language remains diagnostic as a specimen.)
5+ years managing 5-10 product designers. "Ship quickly while maintaining craft." The verb is ship. The constraint is quality. "First-principles thinking about UI patterns built for a prior era" sounds like pattern invention until you read the context: questioning existing patterns in service of shipping better product, not conducting research into what new patterns should be. Applied design leadership at the most consequential AI company in the world. The AI context raises the bar. The job is still leading designers who ship.
The sharpest comparison is within OpenAI itself. Their Model Designer role sits in Applied AI. The posting says the "model is the product itself." Responsibilities: working with researchers to understand, predict, and design model behavior. Defining new human-AI interaction paradigms. Same company, different role, entirely different hiring problem. The Model Designer is lab-native. The Product Design Manager is applied. If the five signals can't sort these two, they're broken. They sort cleanly.
Sorting Juno's Targets
Applied Trust and Delegation Design — Lead Hard
Abridge. High confidence. Every signal points here. Their Senior Product Designer posting centers clinical workflows, cognitive load reduction, and "transforming clinical workflows and AI capabilities into intuitive experiences." Their platform emphasizes Linked Evidence and auditable AI that maps summaries to ground truth for trust and verification. Their Staff Mixed Methods Researcher posting lists "AI-assisted workflows, trust, quality, human review" as a bonus area. Consequential-delegation design in a regulated domain. The human-system operating contract running through Juno's portfolio is the exact problem Abridge is hiring for.
Lead with: Thermo Fisher. Specifically the regulated-domain trust architecture: designing the system that determined when a scientist trusted an instrument's AI-generated recommendation versus overriding it. The verification and auditability problems are structurally identical to Abridge's Linked Evidence challenge. Then Carrier IQ from the Agentic Labs portfolio at junochen.com. The oversight architecture for a logistics agent mirrors the clinical-review workflow Abridge needs: AI generates a structured output, a human with domain expertise decides whether to accept, modify, or reject it. TinyFish adds current currency. Enterprise agent deployment with governance challenges proves Juno is inside this problem today, not citing it from two years ago. "Trust Is the New Interface" maps directly to Abridge's trust-and-verification language.
Suno, Head of Product Design. High confidence for category, moderate confidence for fit. The posting is design leadership: leading and scaling Product Design, Design Engineering, and UX Research. Consumer technology at scale. People leadership plus creative leadership plus product strategy. Suno says explicitly:
"AI is our tool, not our identity."
That sentence alone sorts them. No lab-native invention required. AI fluency, yes. AI research capability, no.
Lead with: Design-org leadership track record and consumer product craft. For the AI dimension, Brand Pulse from the Agentic Labs portfolio. A brand-monitoring agent in a creative-tools context is the closest analog to Suno's "AI as creative tool" positioning. TinyFish proves current AI-product velocity: rapid enterprise agent deployment demonstrates the speed Suno's consumer-scale posting demands. Note: still a Watch-tier company for other reasons (level verification, buyer access). The category sort doesn't change the actionability tier.
OpenAI, Product Design Manager. High confidence for category. Use with caution on actionability given the posting may no longer be live. If a similar role reappears, the positioning is design leadership with systems thinking across consumer and enterprise surfaces.
Lead with: Alibaba ecosystem complexity (cross-surface systems design at scale) and the team-leadership track record. TinyFish adds AI-product context without overweighting it. For the "UI patterns built for a prior era" language, Retail Velocity from Agentic Labs is the best evidence: designing interaction patterns for an agent-driven retail operations workflow where the prior-era pattern (dashboard monitoring) had to be rethought for an agentic model.
Lab-Native Pattern Invention — Genuine Mismatch or Compensating Strength
Anthropic, Design Engineer, Education Labs. Genuine mismatch on multiple dimensions. IC, not leadership. Embedded in a research team, not leading a design function. Requires published work or research methodology in learning sciences or HCI. The design material is human capability development with AI, not trust architecture for AI output. The posting explicitly says it's "likely not right for someone seeking immediate engineering management or a large-team lead role."
Honest assessment: This role wants someone who has invented interaction patterns from inside a research context and can point to published or shipped paradigm work. Juno's Agentic Labs work is applied product design, not paradigm research. Deprioritize unless a different Anthropic posting opens closer to applied design leadership.
OpenAI, Model Designer. Genuine mismatch. The design material is the model itself. The role works with researchers to shape model behavior, not to design interfaces around model output. "Defining new human-AI interaction paradigms" here means influencing what the model does, not what the user sees. The posting centers "balancing capability, inferred user intent, and user trust" from the model side, and developing data-collection strategies.
Honest assessment: Juno designs the human layer around AI systems. This role designs the AI layer that humans interact with. The direction of influence is reversed. A scope mismatch, not a positioning problem.
Anthropic, Research Product Manager, Labs. Moderate confidence, hybrid. The posting owns "moonshot products" and leads 0-to-1 development from research to shipped products. Lab-native elements (research translation, prototype-driven) coexist with applied elements (shipped products, product development). Compensating strength might suffice here if Juno can demonstrate research-to-product translation concretely. Strongest evidence: Carrier IQ from Agentic Labs. That system required translating research findings about trust thresholds in logistics workflows into a shipped product with specific escalation and override patterns. The movement from "we learned that carriers override at this threshold" to "here's the shipped intervention architecture" is research-to-product translation, even if the research context was applied rather than academic. Worth a conversation, not a hard pass. Not a lead-with target.
The Honest Gap, Stated Plainly
For lab-native roles, the grain of truth is more specific than "hasn't worked at a frontier lab." Three things Juno has not done:
- Shaped model behavior as a design decision. Her work designs the human response to model output, not the model output itself.
- Published research on human-AI interaction paradigms.
- Worked embedded in a research team where the design artifact is a study finding or capability framework rather than a shipped interface.
Better framing does not close these gaps. They are experience gaps, not positioning gaps. The compensating strength is real: Juno has designed trust, delegation, and control architectures for AI systems entering high-stakes workflows, which is a problem most lab-native designers haven't touched. But compensating strength and core requirement are different things. Conflating them in an interview is how you lose credibility in the first ten minutes.
Applying the Signals to Your Watch List
For companies not yet posting design leadership roles but on Juno's target list, the five signals work the moment a posting appears. Rubrik and Headway both sit in the space where AI and complex systems shift users into supervising agent-driven work. If either posts a design leadership role, expect it to sort applied: the design problem is supervision, exceptions, controls, and recovery paths, not model-behavior invention. Same logic applies to Ramp, Brex, Amplitude, and Vanta. These companies need trust architecture for AI entering consequential workflows. Run the five signals to confirm when their postings land, but the company-level context already points applied.
The only watch-list companies where a lab-native posting could plausibly appear are those building their own models or running internal research teams. For most of Juno's target universe, applied-trust positioning is the right one. The sorting framework matters most at the frontier labs (Anthropic, OpenAI) where both role types coexist under the same roof.
When the Question Comes Up
"Your AI work seems more applied than foundational."
For applied-trust roles (Abridge, Suno leadership, enterprise AI): "That's accurate, and it's the point." Follow with: "The hardest design problem in AI right now is designing the trust architecture that determines whether a human delegates a consequential decision to an agent or overrides it. That's what I built in the Carrier IQ and Brand Pulse systems at Agentic Labs and what I designed the oversight architecture for at Thermo Fisher." High confidence.
For lab-native roles where you're in the conversation anyway: "I design the human side of the delegation contract, not the model side." Follow with: "I know what happens when the trust architecture fails in a regulated workflow. I know what 'override' looks like when the stakes are clinical or financial. That's a different expertise than paradigm invention, and I think you need both on this team." Moderate confidence. Works only if the hiring manager already sees a gap in their team and is open to complementary expertise.
For roles that are clearly model-as-material: Don't reframe. Ask instead. "I want to understand how much of this role is shaping model behavior versus designing the human layer around it, because my strength is the second and I want to be honest about that." Use with caution. Only when you've already decided the role is a stretch and you're optimizing for relationship rather than conversion.
Quick-Reference Sorting Table
Pull this up before outreach or interview prep.
| Company | Role | Category | Confidence | Lead With | TinyFish Use | Honest Assessment |
|---|---|---|---|---|---|---|
| Abridge | Senior/Staff Product Designer | Applied trust | High | Thermo Fisher (regulated trust architecture) + Carrier IQ (agent oversight in consequential workflow) + "Trust Is the New Interface" | Enterprise agent governance proves current AI currency | Strongest category match on the list. Consequential delegation in regulated domain is Juno's exact positioning. |
| Suno | Head of Product Design | Applied trust | High (category) / Moderate (fit) | Brand Pulse (creative-tools agent) + design-org leadership track | Rapid agent deployment proves current shipping fluency | Category match is clean. Actionability depends on level verification and buyer access. Watch tier. |
| OpenAI | Product Design Manager | Applied trust | High (category) / Use with caution (actionability) | Alibaba (cross-surface systems at scale) + Retail Velocity (rethinking prior-era patterns for agentic model) | Current AI-product context | Posting may be filled or paused. If similar role reappears, positioning is strong. |
| Anthropic | Design Engineer, Education Labs | Lab-native | High | N/A | N/A | Genuine mismatch. IC research role requiring published HCI/learning-science work. Deprioritize. |
| OpenAI | Model Designer | Lab-native | High | N/A | N/A | Genuine mismatch. Model is the design material. Direction of influence is reversed. |
| Anthropic | Research PM, Labs | Hybrid | Moderate | Carrier IQ (research-to-product translation in trust thresholds) | Light context only | Compensating strength may suffice. Worth a conversation, not a lead-with target. |
| Rubrik, Headway, Ramp, Brex, Amplitude, Vanta | When posted | Expected: applied trust | Moderate (pre-posting) | Run five signals to confirm; default to consequential-delegation positioning | Enterprise agent deployment as current proof | Company-level context points applied. Confirm with posting language when live. |
The Sorting Habit
Every AI design posting from here forward: run the five signals. Where does research appear. What's the design material. What verbs dominate. What's the success metric. What's absent. Three of five gives you a category. The category tells you which version of your story to tell, or whether to tell it at all.
The strongest position you can walk into a room with is knowing exactly which AI design problem you solve and saying so in the first thirty seconds. Including when the answer is "not this one."
- Abridge's trust-research hiring: Their new Staff Mixed Methods Researcher posting explicitly names "AI-assisted workflows, trust, quality, human review" as evaluation areas, which signals that Abridge is building internal research capacity around the exact trust-and-verification problems this dossier maps to Juno's strength.
- Overreliance research complicates "more explanation": Chen, Liao, Vaughan, and Bansal found that feature-based explanations increased overreliance rather than improving outcomes, which means Juno's trust-architecture positioning should emphasize appropriate reliance and verification cost, not transparency for its own sake.
- EU AI Act override language: Article 14 of the EU AI Act requires that high-risk AI systems enable humans to monitor, interpret, override, reverse, and interrupt operation, giving Juno concrete regulatory vocabulary for applied-trust interviews at Abridge, Capital One, or any company shipping into regulated workflows.
- Anthropic's Research PM, Labs: The moonshot-products posting blends lab-native and applied elements in ways that make it the most ambiguous sort on Juno's list, worth monitoring for language revisions that would shift it more clearly into one category.