The guideline references Domain 1 (Inference-Aware UX) and asks for mapping to Juno's artifact domains. Those domain names don't appear on the current junochen.com homepage, and the lab case routes that would organize them still return 404. These sheets map directly to the standalone Lab apps and published work.
Stripe — Senior Staff Product Designer, Data & AI
Scope the role before you position for it
Stripe's OpenRouter acquisition is the loudest AI signal in their public posture. The natural assumption is that a "Data & AI" design role sits inside that story. Read the actual posting. It covers Data Products plus Ecosystems and Startups — Sigma, Data Pipeline, Atlas, App Marketplace, Business Network. No mention of OpenRouter, inference routing, Token Billing, Link, agent wallets, or Agentic Commerce. The parallel Product Lead posting confirms the mandate: merchant-facing analytics, data pipeline, and the growth edge — agentic access to merchant business data over the next two to three years.
The design problem you'd own: how do merchants maintain trust, visibility, and control when AI agents query and act on their business data? That includes natural-language querying (currently in private and public preview), automated reporting, and the analytics surfaces inside the Stripe Dashboard.
(The prior Stripe sheet in Issue #11 addressed the Link role and centered on spending authority. Different role, different evidence selection.)
Confidence levels on scope:
- High: Analytics, data pipeline, ecosystem, and startup surfaces.
- Moderate: AI-native merchant data access and agent-facing analysis — the Product Lead posting places these in the Data Products roadmap explicitly.
- Speculative: Anything involving OpenRouter integration, inference routing, or agent payment infrastructure. No public source assigns those to this team.
If you walk in talking about inference-aware UX and model routing, you may be answering a question nobody on this panel is asking.
Lead with CarrierIQ
This role's design problem is provenance and oversight. When an agent queries a merchant's revenue data, generates a report, or triggers an action based on financial metrics, the merchant needs to see what was accessed, what the agent concluded, whether that conclusion is correct, and what happens if it isn't. CarrierIQ demonstrates this cycle: structured intake, operating stages, evidence with attribution, operator notes, re-verification, approval gate. That approval gate matters here. It shows you've designed the moment where a human decides whether to accept or reject an automated system's output based on inspectable evidence.
"I've designed the interaction pattern for human oversight of automated data processes — here's what that looks like when the stakes are high and the evidence needs to be traceable."
Supporting evidence
- Brand Pulse as the analytics-synthesis companion. Source-agent status, source-labeled evidence, and synthesis with attribution map to the AI-generated analytics surfaces Stripe is building. Use it to show you've thought about how AI-generated insights should present their sources and confidence.
- Trust essay for the conceptual frame — specifically the Watch / Verify / Delegate vocabulary. This gives the panel language for how merchant oversight should evolve as trust in agent-generated analytics increases. Don't lead with it. Use it to deepen a conversation that CarrierIQ starts.
- Alibaba (Head of Design and Research, North America) as production proof. B2B marketplace redesign at $50B+ GMV is the closest thing in your portfolio to Stripe's scale and commercial domain. It confirms you've operated inside this kind of complex, high-volume merchant ecosystem. The Lab artifacts differentiate you. Alibaba confirms you can operate at this weight class.
Your differentiator
At this level, every serious candidate has enterprise product experience. Some will have designed dashboards and analytics surfaces. What they almost certainly won't have is a working artifact that demonstrates the approval and re-verification layer for agent-mediated data access — the interaction where a human reviews what an automated system did with their data and decides whether to trust it. CarrierIQ is that artifact. The Claims Ledger specification from Issue #12 extends this — consequential statuses carrying evidence, owner, timestamp, explicit unknowns — but lead with the working app, not the spec.
What NOT to lead with
- The inference-routing narrative. Speculative for this role. If the panel brings up OpenRouter, respond to it. Don't anchor your positioning there.
- Retail Velocity. Visibly names TinyFish dependencies. Provenance questions will follow, and those questions spend interviewer attention on attribution rather than your design judgment.
- Delegation Contract vocabulary (Proposed / Staged / Authorized / Committed / Settled / Compensating). Specified, not shipped. Stripe will want to see demonstrated judgment.
Org-structure note
Speculative. The posting names no hiring manager, team size, or reporting line. The Product Lead posting sits under Revenue & Financial Automation. No public "I'm hiring" post from a Data Products design manager surfaced. Treat any org-structure claims from recruiters as unverified until you confirm them in conversation.
Lab case routes on junochen.com still return 404. The standalone apps work. Any outreach or follow-up referencing CarrierIQ or Brand Pulse requires the subdomain URLs above. The homepage does not surface them.
Anthropic — Product Designer, Safeguards
The design surface is the customer's review workflow
Enterprise Frontier Safeguards is a monitoring architecture. Here's what Anthropic has published about how it works: automated systems inspect a rolling window of customer traffic for serious misuse signals. Detected flags go to the customer. Customer personnel conduct human review by default — not Anthropic staff. Activity data can remain in the customer's own cloud account under their encryption keys and access policies.
What Anthropic has not published: a flag-review dashboard, an investigation console, a configuration API, an escalation workflow, or a reviewer interaction model. The announcement describes operating actors and control categories but shows nothing about what the customer's security or compliance team actually sees and does when a flag arrives.
That's the design problem this role likely owns. It sits at two trust levels: Anthropic trusting customers to conduct competent review, and customers trusting that automated detection is surfacing real signals with enough evidence to act on.
A secondary surface worth preparing for: Claude Fable's model page describes flagged queries being automatically routed to less capable models, with customers not charged the higher-tier price. That's model-level safeguarding, distinct from EFS's monitoring and review architecture. But it creates its own design problem: a user's output quality degrades mid-session without a clear explanation of why, what changed, or what recourse exists. If the Safeguards designer owns both the review workflow and the model-fallback experience, the role spans two different trust failures — one where a human reviewer needs enough evidence to act, and one where an end user needs enough transparency to understand what just happened to their session. You don't need to lead with this. But if the panel raises model routing, you can speak to the experience cost of silent capability changes — the Issue #9 analysis gives you the frame.
Confidence levels:
- High: Customer teams control access and conduct human review by default. EFS is in phased rollout.
- Moderate: EFS creates a substantial customer-side investigation workflow — flags need evidence, false positives need clearance, reviewers need to understand why something was flagged.
- Speculative: Whether the Product Designer, Safeguards role owns EFS specifically, or whether it also covers model-fallback experience.
Lead with CarrierIQ, framed for review workflows
carrier-iq.junochen.com — same artifact, framed differently than Stripe.
For Anthropic, the relevant pattern is: automated system detects something, presents evidence to a human reviewer, reviewer inspects and makes a disposition decision, system records the outcome. CarrierIQ demonstrates every step. The approval gate here isn't about merchant oversight of business data — it's about a security or compliance analyst reviewing a misuse flag with enough structured evidence to confirm or clear it.
CarrierIQ's re-verification step is particularly relevant. A previous determination gets revisited when new evidence appears. False-positive management is the operational reality of any automated detection system, and you have a working artifact that shows how re-review should work.
"I've designed the interaction where a human reviewer decides whether an automated detection is real, based on structured evidence with clear provenance. That's the core loop for customer-side safeguard review."
Supporting evidence
- Trust essay — the five handoffs and Watch / Verify / Delegate vocabulary map directly to EFS's escalation model. Automated detection is the Watch layer. Customer review is Verify. The question of when Anthropic's own team gets involved is the Delegate boundary. Use it to show you've thought about where human authority sits in a layered safeguard system.
- Brand Pulse as secondary evidence for source-labeled attribution. When a flag arrives, the reviewer needs to see what triggered it, from which traffic, with what confidence. Brand Pulse's source-agent status and evidence attribution patterns apply here, though less directly than CarrierIQ's full review cycle.
- Thermo Fisher (Product Design Director, BCG Digital Ventures) if the conversation turns to regulated-industry experience. Healthcare and life sciences carry compliance and safety review requirements that parallel Anthropic's safeguard context. Background proof, not a lead.
Your differentiator
The competitive set for this role will skew toward consumer AI product experience or enterprise security backgrounds. What most of them won't bring is the specific interaction CarrierIQ demonstrates: a reviewer receiving an automated flag, inspecting structured evidence with visible provenance, making a disposition decision the system records, and returning to a previous determination when new evidence arrives. Most enterprise security tools stop at surfacing what happened. CarrierIQ shows a reviewer deciding what to do about it — evidence basis visible, decision recorded, re-verification built in.
What NOT to lead with
- Enterprise scale (Alibaba). The Safeguards problem is about trust architecture and review workflows. Scale credentials don't differentiate here.
- Retail Velocity. Same TinyFish provenance issue as Stripe, and the confidence-graded synthesis pattern is less relevant to a flag-review workflow than CarrierIQ's approval gate.
- Framework vocabulary. "Delegation Contract," "Inference Conflict Surface," and similar labels describe specified-not-shipped work. Anthropic's research culture will probe whether you've built the thing or named the thing. Lead with the built thing.
Same constraint as Stripe. Lab case routes return 404. Use subdomain URLs directly. The homepage does not surface CarrierIQ, Brand Pulse, or the Trust essay's connection to safeguard design. You'll need to build that bridge yourself in outreach or conversation.
- Anthropic's Project Swap study: The 24 September research found that 85% of the gap between observed and optimal outcomes came from the agent misrepresenting preferences, not from bargaining failure — directly relevant if Anthropic's Safeguards panel asks how you'd validate that an automated system understood intent before acting on it.
- NIST agent identity implementation: NIST announced on 24 September that its DevSecOps Build 3 will scope agent identification, authentication, and authorization inside the software-development lifecycle, which may give both Stripe and Anthropic panels a concrete policy reference point for agent authority questions.
- Causal explanation evidence: A September Nature paper on autonomous vehicles found that causally grounded concept explanations improved mental models for 8 of 9 experts, but only during surprising events — useful framing if either panel asks how you'd design explanation for flagged or unexpected agent behavior.
- OpenAI's Identity role as a signal: The Product Designer, Identity posting asks the designer to define human-to-agent and agent-to-agent authorization patterns, which confirms that identity and authority are becoming standalone design surfaces across multiple companies, not just Anthropic's Safeguards concern.
