Version: 7 · Date: September 12, 2026 · Author: Reva Sandoval Purpose: Self-contained portfolio reference for AI systems preparing outreach, cover letters, and interview responses on Juno's behalf. Supersedes v6.
Identity and Status
Juno Chen. Bay Area. Design leader for AI-native platforms. Homepage positioning line:
"I build design teams and ship products where the system is complex, the stakes are real, and the design has to hold."
Most recently at TinyFish (past tense, always). Prior: founding design leader at BCG Digital Ventures, shipping Thermo Fisher, American Red Cross, Equinox+, and Allē. Head of Design & Research at Alibaba.com North America, growing the design team from six to twelve-plus.
Live screening risk: A public TinyFish/Coca-Cola case page now exists (last modified September 12). The homepage still labels TinyFish "NDA protected" while linking to this page. Anyone who clicks through sees the contradiction. See Positioning Rules for handling.
Timing context: Post-Labor Day hiring reactivation is underway. Three new OpenAI design postings appeared September 10. Capital One posted three design-manager roles September 11. Outreach sent this week lands in reactivated inboxes.
Three Concepts That Extend the Trust Essay
The trust essay uses the scalar Watch → Verify → Delegate ladder and five handoffs. These three concepts extend that framework. In any conversation, ground each one in production evidence before introducing the concept name.
Authority-as-vector. The trust ladder implies a user sits on one rung. Production says otherwise. A procurement manager on Alibaba fully delegates supplier search while requiring manual approval for purchase orders above a threshold. Authority varies by action type and stakes level simultaneously — it's a vector, not a position on a scale. The Alibaba redesign demonstrates this implicitly: browsing (low-stakes, high autonomy) was separated from RFQ submission (high-stakes, manual gate). The TinyFish case demonstrates it explicitly: intervention roles (None, Verify, Wait, Adjudicate, Teach, Approve) vary by claim state, not by user trust level.
Commitment sequence. Five states for consequential agent actions: propose → stage → authorize → commit → settle. This decomposes the trust essay's Decision Gate handoff into temporal phases. Propose: the agent surfaces an intended action with evidence. Stage: the action is prepared but not executed — reversible at zero cost. Authorize: the user grants permission, scoped to this action. Commit: the action executes. Settle: downstream effects resolve and the system confirms final state. Carrier IQ's approval flow covers propose through commit. TinyFish's humanReview[] object covers propose through authorize. Neither demonstrates the full settle phase with verified downstream resolution.
Delegation Contract. The lead forward-looking artifact. A Delegation Contract is the interface that makes authority-as-vector and commitment sequence visible and adjustable. It shows: what the agent may do without asking (per action type and stakes level), where the commitment sequence requires user presence, what happens when the user is absent at an authorization point, and how authority allocation shifts based on accumulated evidence. The Delegation Envelope specifications from Issue #9 — Confidence/Authority Matrix, Revocation Sequence, Contract-Variance Disclosure — are its direct precursor. The Delegation Contract unifies those three into a single user-facing surface. Authorization is a negotiated, adjustable relationship between user and agent, and the user needs one place to see and modify that relationship. It is not built. Detailed visual specification lives in the Artifact Build Queue (separate deliverable). Route through verbal positioning and the precursor specs, not through a portfolio link.
Forward-Looking Artifact Inventory
Four domains. Status reflects what is publicly inspectable as of September 12.
| Artifact | Domain | Status | Grounded by |
|---|---|---|---|
| Delegation Contract | Human-agent system design | Specified, not built. Precursor specs in Issue #9. | Carrier IQ approval flow, TinyFish intervention roles, Alibaba procurement gates |
| Correction Lineage | Inference-aware UX | Partially visible. TinyFish case shows humanReview[] with receipt and evaluation-promotion. Does not show verified behavior change. | TinyFish promotedToEval field, correction claims framework |
| Attention-Load Analysis | Intent-based interaction | Specified in Issue #10. Not built. | Brand Pulse parallel-agent visibility, Carrier IQ multi-carrier session tracking |
| Safe Departure / Bounded Re-Entry | Agent infrastructure as UX | Specified in Issue #10. Not built. | TinyFish stop conditions (three enforced), Red Cross supervisor-approval gates |
The Delegation Contract is the artifact most relevant to the OpenAI Identity role and to any company building agent authorization. It remains unbuilt. This is the single largest gap between Juno's verbal positioning and inspectable evidence.
Agentic Labs Foundations
Three live applications. None linked from the homepage's Selected Work section. All publicly reachable. Link directly to the relevant app in outreach when it grounds a specific claim — do not assume the screener has found them.
Carrier IQ. Insurance quoting across eight carriers. Structured intent capture, parallel execution visibility (Session → Navigate → Fill → Extract → Verify), graduated review states (Bindable, Normalize, Referral, Call review), evidence-backed approval with carrier trust signals. Grounds the Delegation Contract's propose-through-authorize sequence and the Attention-Load Analysis (multiple carrier workflows competing for user attention simultaneously).
Brand Pulse. Competitive brand intelligence from Reddit/X. Source-labeled evidence cards, sentiment and theme classification, parallel agent search with live scoring, agent trace. Grounds Correction Lineage — source-labeled evidence is the precondition for traceable correction, because you cannot correct a claim you cannot trace to its source. Also grounds the Attention-Load Analysis through parallel-agent search with live scoring. Does not demonstrate authorization or commitment sequence.
Retail Velocity. Restaurant beverage-displacement opportunity scanner. Confidence-labeled ranked results, evidence snippets, agent log. Grounds the Delegation Contract's evidence-display requirements — ranked opportunities with confidence labels are the information layer a user needs before authorizing action. The connection is thinner than Carrier IQ's; Retail Velocity shows evidence display without an authorization step. Note: the page visibly credits "TinyFish Search · TinyFish Fetch · TinyFish Agents." Do not use this branding as evidence of a current TinyFish relationship.
Evidence gap across all three Labs apps: None show a correction-to-next-run sequence, a settle phase, or authority-as-vector allocation. The TinyFish case page partially fills the correction gap with humanReview[], but correction does not yet rewrite the restaurant record, and two distinct real logs remain unmerged.
Case Studies
All metrics are Juno's published claims, not independently verified.
Alibaba.com. Head of Design & Research, North America. $50B+ GMV platform, 200K+ suppliers, 25M+ desktop sessions. Team 6 → 12+. Led end-to-end procurement redesign across homepage, search, product detail, and onboarding. Results: +7% DAU, +20% transactions, +2.2-point NPS, 47% fewer security concerns. Authority-as-vector is visible in the design: browsing trust signals (verified suppliers, Trade Assurance, supplier tenure) were separated from transaction commitment gates (MOQ, inspection reports, Ready-to-Ship vs. Custom Order). Use as enterprise-scale evidence and production precedent for multi-stakeholder trust design.
TinyFish/Coca-Cola. Agentic system for restaurant intelligence. 5,505 Pennsylvania restaurants scanned, 1,386 flagged for human judgment, 163 design decisions on real data, 3 enforced stop conditions. Claim states: decision-ready, partial/degraded, waiting/conflicted, unknown. Intervention roles: None, Verify, Wait, Adjudicate, Teach, Approve. The case explicitly labels 1 of 186 traces as real, 185 as reconstructed from final-state fields, and run histories as derived. Three shipped fixes: binary number presented as confidence, fabricated summary copy, unreadable truncated JSON. Strongest agentic-design evidence in the portfolio. See Positioning Rules for NDA-conflict handling.
Thermo Fisher mySupply. Product Design Director, BCG DV. 0→1 supply-chain platform, six pharmaceutical partners, nine sites. $20M margin opportunity, 100% partner adoption, 83% IRR. Commitment sequence in a non-AI context: batch release required QA authorization before shipment (authorize → commit → settle with downstream partner confirmation).
Allē. Product Design Director, BCG DV. 30M+ members, 40K+ practices. 3.2× redemption, 47% reactivation, CAC reduced from $92 to $42. One loyalty model serving both members and providers.
American Red Cross. Product Design Director/GM, BCG DV. National disaster-relief platform replacing six legacy systems. $847K disbursed, 1,689 cases in first two weeks. Commitment sequence with real consequences: eligibility determination → supervisor approval → payment release → FEMA compliance.
Equinox+. Product Design Director, BCG DV. 600K+ premium members, 200+ clubs. 0→1 in 90 days, 4.8-star App Store rating. Five brands on one platform with shared token architecture.
Trust Essay Mapping
The live essay defines five handoffs: Intent-Setting, In-Progress, Output Review, Decision Gate, Loop Feedback.
How authority-as-vector extends the ladder: The ladder assumes uniform progression. Authority-as-vector says a user may be at "Delegate" for data gathering and "Watch" for financial commitment simultaneously. In interviews, acknowledge the ladder as the starting framework, then introduce the vector concept as what production taught — TinyFish intervention roles that vary by claim state, not by user trust level.
How commitment sequence extends the Decision Gate: The essay groups consequential authorization into one handoff. The commitment sequence decomposes it into five temporal phases. Thermo Fisher already hints at this in the essay (continuous automated monitoring vs. human QA release decision). Use Thermo Fisher as the bridge.
| Handoff | Case evidence | Agentic Labs evidence | Artifact domain |
|---|---|---|---|
| Intent-Setting | Alibaba RFQ structure, Carrier IQ profile capture | Carrier IQ, Brand Pulse launch surfaces | Intent-based interaction |
| In-Progress | TinyFish parallel workstreams | Carrier IQ session stages, Brand Pulse live scoring | Agent infrastructure as UX |
| Output Review | TinyFish claim states, evidence provenance | Brand Pulse source-labeled cards, Retail Velocity confidence labels | Inference-aware UX |
| Decision Gate | Red Cross supervisor approval, Thermo Fisher QA release | Carrier IQ Approve Bind | Human-agent system design |
| Loop Feedback | TinyFish humanReview[] with promotedToEval | Not yet demonstrated | Inference-aware UX |
Company Positioning
Ranked by proof readiness, not excitement. The last explicit tier document is v5 from August 30. Gusto's Head of Design role is retired.
Act Tier
1. OpenAI — Product Designer, Identity. $245K–$310K + equity. This role is designing the authorization and permission layer for agent actions. Lead with the Delegation Contract concept. The commitment sequence maps directly: when does an agent need to confirm identity before committing? Authority-as-vector maps to the permissions model: different agents get different authority levels for different resource types. Production proof: TinyFish intervention roles, Carrier IQ approval flow, Red Cross identity-and-eligibility checks. Ask early: whether "identity" means user-facing authentication UX or the deeper question of how agents prove identity to each other and to external services. The answer changes which case to lead with. Three new OpenAI design roles posted September 10 (Accessibility, Codex, Youth) — none manager+, but the Codex posting shares delegation vocabulary with Identity. Monitor for overlap.
2. Vanta — Head of Design. $365K–$513K + equity. Remote US. ~40-person org. Lead with authority-as-vector applied to compliance: Vanta's product determines what an automated compliance agent may certify vs. what requires human auditor sign-off, and that boundary shifts as customer trust in automation grows. The Delegation Contract is the artifact that makes this boundary visible. Production proof: Thermo Fisher compliance workflows, Alibaba procurement trust signals, Red Cross FEMA-compliance views. At this comp range with a 40-person org, this is a true leadership role. If the posting doesn't mention hiring and team building, probe.
3. Altana — Head of Product Design. $240K–$300K. Brooklyn. Lead with Correction Lineage. Altana's supply-chain intelligence product requires users to trust AI-generated risk assessments, and that trust depends on whether the system incorporated their prior corrections. Did the risk model learn from the last false positive the user flagged? That's the Correction Lineage problem — receipt, incorporation, verified improvement. TinyFish's humanReview[] with promotedToEval is the closest production evidence. Output Review (the trust-essay handoff) provides supporting structure: source-labeled evidence so users can evaluate the assessment before acting. Production proof: TinyFish correction receipt, Brand Pulse source-labeled evidence cards. September 4 positioning still holds for framing; update the lead artifact.
4. Amplitude — Head of Product Design. $317K–$476K total target cash, Bay Area. 15 designers. Lead with authority-as-vector applied to analytics: a product-analytics platform whose users range from data scientists to marketing managers needs different default authority levels for different user expertise and decision stakes. How much should the system automate insight surfacing versus requiring analyst confirmation? The Delegation Contract concept applies directly to this calibration. Production proof: TinyFish confidence-display fixes (presenting statistical confidence legibly), Brand Pulse sentiment scoring. Team-scaling evidence: Alibaba 6 → 12+ during a platform redesign. Probe whether the role owns design strategy or executes product-management strategy.
5. Headway — Staff Product Designer, Provider Practice Growth. $212K–$265K + equity. Title changed from "Provider CRM & Prospect Management" — the reframe from tooling to outcomes is the signal. Lead with Allē's dual-audience loyalty model. Production proof: Allē 3.2× redemption and CAC reduction. The Delegation Contract and agentic concepts are less relevant here unless the role involves AI-assisted provider matching.
Watch Tier
OpenAI — Product Design Lead, Growth–Codex. $347K–$405K + equity. Manages 2–5+ designers, hands-on. Secondary OpenAI target if Identity doesn't progress. Do not pursue simultaneously.
Capital One — Director, Product Design, AI in Experience Design. $251K–$287K (SF/NY). Posted August 21. Three new design-manager roles posted September 11 confirm active org investment. Authority-as-vector applies to financial AI. Production proof: Alibaba enterprise scale, Red Cross financial disbursement gates.
Slack — VP, Product Design / Principal Architect. Posted September 2. IC role despite VP title (the "Principal Architect" qualifier). Monitor for scope clarification. If truly a systems-design IC role, the Delegation Contract concept and Equinox+ multi-brand architecture are relevant.
Handshake — Staff Product Designer, Student Experience. Posted September 4. Below Juno's typical scope unless the role carries outsized influence. Monitor.
Gusto. Head of Design is retired. Senior Product Design Manager, Banking remains live but below target scope. Move to retired unless a Director+ role appears.
Positioning Rules
TinyFish is past tense. The new public case page changes what is inspectable but does not change this rule. In outreach: reference the agentic-design patterns and what they taught, not the client name. In interviews: if asked directly, the case page is citable, but lead with the design problems solved (confidence display, evidence provenance, intervention-role design).
NDA-conflict handling. The homepage says "NDA protected" while linking to a detailed public page. Until this is resolved on the site, do not proactively send the case-page URL in outreach. If a screener finds it through the homepage link, it is public and discussable. Do not pretend the conflict doesn't exist if asked.
BCG DV attribution. Thermo Fisher, Red Cross, Equinox+, and Allē were shipped as a founding design leader at BCG Digital Ventures. Individual case pages don't repeat this. Include the attribution in any written positioning that references these cases.
Agentic Labs visibility. Carrier IQ, Brand Pulse, and Retail Velocity are not linked from the homepage. Link directly to the relevant app in outreach when it grounds a specific claim.
Evidence provenance in the TinyFish case. The case page labels 1 of 186 traces as real, 185 as reconstructed, run histories as derived. This honesty is a strength. Do not overstate the correction chain: humanReview[] shows receipt and evaluation-promotion, not verified behavior change.
Prohibited:
- Do not claim the Delegation Contract is built.
- Do not claim the correction chain is complete.
- Do not present TinyFish as a current engagement.
- Do not claim independent verification of case-study metrics.
Interview Questions
"How do you design trust in AI systems?" Start with the five handoffs, then introduce authority-as-vector: trust isn't a single dial. At TinyFish, the same user needed six different intervention postures depending on claim state. A "decision-ready" record needed no human touch. A "waiting/conflicted" record required adjudication. The design job was making that variation legible. Ground with TinyFish intervention roles, Carrier IQ graduated review states. Artifact: Delegation Contract concept (verbal), Carrier IQ approval flow (inspectable).
"Show me a 0→1 product." For enterprise/B2B: Thermo Fisher (12 months, $20M margin, 100% adoption, six pharmaceutical partners). For consumer/platform: Equinox+ (90 days, 4.8 stars, five brands on one platform). Both demonstrate commitment sequences in production.
"How do you think about AI agents taking consequential actions?" Lead with the commitment sequence: propose, stage, authorize, commit, settle. The key design question is what happens between stage and authorize — the user needs to see what will happen, confirm they want it, and know what "settle" looks like before they say yes. Ground with TinyFish stop conditions, Red Cross payment-release gates, Thermo Fisher QA holds.
"What's your experience scaling a design team?" Alibaba: 6 → 12+ while leading a platform redesign across four major surfaces. Sequenced hiring to delivery sprints — researchers first to build procurement-behavior evidence, then designers to execute against validated problems. BCG DV: built design teams for four 0→1 products across healthcare, fitness, financial services, and disaster relief.
"How do you handle disagreements with product or engineering leadership?" Alibaba: identified the mismatch between consumer-style browsing and professional procurement before any screen redesign began. Built the research case, presented to executives, secured the mandate. The disagreement was structural — the product's interaction model was wrong for its users — and the research made that visible. Red Cross: the design argument was that six systems serving the same case was the problem, not six bad interfaces.
"What would you build here that you haven't built before?" For OpenAI Identity: the Delegation Contract. The unified surface that lets a user see and adjust what an agent may do across action types and stakes levels. The precursor work exists — TinyFish intervention roles, Carrier IQ approval flows, the three Delegation Envelope specifications — but the unified artifact has not been built. That is what this role exists to solve.
-
Meta Muse's approval architecture: Anthropic's incident assessment found a monitoring model accepted biased reasoning from the acting model it was supposed to check — a finding that directly pressures Muse's Sentinel-based approval design and the Delegation Contract's assumptions about independent verification.
-
OpenAI Agents API settlement gap: The session documentation describes cancelling an active turn while retaining prior work, but does not address reversal of external effects already dispatched before cancellation — the exact settle-phase problem the commitment sequence is designed to make visible.
-
Broken canvas link on TinyFish case: The new case page's "Explore the Canvas" link targets
cs-06-coke-intelligence-teaser.html, which returned 404 during the September 12 check — a routing defect that undermines the page's strongest evidence before a screener reaches it. -
Anthropic role status change: The previously tracked Core Apps design URL now redirects to an error state, which could mean filled, cancelled, or renamed — worth monitoring for a replacement posting that might carry authorization-design language similar to OpenAI Identity.