Verified: August 22, 2026 Surfaces checked: junochen.com homepage, Trust essay, Carrier IQ app and case page, Brand Pulse case page and app, Retail Velocity case page and app
This file is the single source for AI-generated outreach, cover letters, and interview prep. Every claim must trace to what is published at junochen.com today or confirmed in Juno's experience. If it is not here, do not generate it.
Identity
Juno Chen — "Design Leadership for High-Stakes Systems" Homepage reads: "Most recently at TinyFish, I shipped AI-native enterprise tools in production." Portfolio: five human-led case studies (Alibaba, Thermo Fisher, American Red Cross, Equinox+, Allē) and four Agentic Work entries on the homepage (Carrier IQ, UAT Sentinel, Retail Velocity, Brand Pulse). Three of the four are authorized for outreach: Carrier IQ, Retail Velocity, and Brand Pulse. UAT Sentinel is not authorized for external reference — do not link, describe, or cite.
TinyFish Rules
TinyFish answers "what have you been doing recently?" and nothing else.
- Past tense only. Currency, not proof.
- No screens, traces, internal artifacts, or links. The pricing essay is publicly accessible but not authorized for outreach.
- Function varies by buyer. AI-native companies: domain credibility. Enterprise: recency signal that needs bridging to prior scaled work. Healthcare/regulated: production-agent context paired with Agentic Labs design proof.
- Thermo Fisher always carries attribution: "work completed as Product Design Director at BCG Digital Ventures."
Artifact Inventory
Three named concepts enter the inventory this cycle. All three are conceptual — no published artifacts. Label them as conceptual in every context.
Three-Claim Framework
Status: conceptual | Domain: Correction Lineage
Products compress three distinct claims into "the agent learned":
- The correction was received.
- The correction changed a defined system object (prompt, policy, memory, evaluator, permission, workflow version).
- The changed system improved behavior under comparable later conditions.
Carrier IQ covers claim 1 and partial claim 2 — evidence, operator notes, re-verification, approval. It does not expose version-level receipts or later-run comparison. Production analogues: OpenAI Presence (proposed changes tested against current version), Vercel's agent design workflow (human-approved rollout after evaluation).
Reversibility Frontier
Status: conceptual | Domain: Task State and Intervention
The point after which a task's next consequential effect can no longer be reliably stopped or undone. Measures progress as remaining recoverable choices, not percentage complete. Separates cancellation requested from cancellation acknowledged across every participant still acting.
The MCP Tasks spec supplies the production mechanism: cancellation communicates intent but does not guarantee the server stops. The Trust essay's Decision Gate retains a human where downstream outcomes are irreversible. The portfolio currently shows no distributed cancellation acknowledgement, side-effect inventory, or last-safe-intervention view.
Inference-Aware UX
Status: conceptual | Domain: Cost, Latency, Reasoning Depth, Uncertainty
How model-effort choices, speed tiers, inference budgets, token costs, timeout fallbacks, and budget-breach decisions shape what users experience. No Juno artifact exposes any of these controls.
Decision Gate asks when human judgment must remain because consequence is irreversible. Inference-Aware UX adds the upstream resource question — whether the system can finish, verify, or escalate before that gate arrives. The link is conceptual. Nothing is published.
Provider context establishing the domain: GPT-5.6 Sol/Terra/Luna effort levels at different price points. Gemini 3.7 Flash low/medium/high thinking levels. Opus 5 effort settings with a Fast mode at roughly 2.5x speed. These are dated provider terms, not stable product constants.
Agentic Labs Foundations → Trust Essay Handoffs
| Foundation | Primary Handoff | What It Demonstrates |
|---|---|---|
| Carrier IQ | Decision Gate, Output Review | Staged execution (Session → Navigate → Fill → Extract → Verify), four review states, evidence attachment, re-verification, operator notes, Approve Bind |
| Brand Pulse | In-Progress, Output Review | Continuous monitoring with scored mentions (sentiment, urgency, source), narrative momentum, platform-coverage visibility |
| Retail Velocity | Loop Feedback | Daily account audit surfacing changes, gaps, and required actions — one run's output informing the next |
| UAT Sentinel | NOT AUTHORIZED | Visible on homepage and publicly responsive. Do not reference, link, or describe in any external communication. |
Case Studies
CS-01 Alibaba. Platform redesign across homepage, search, product-detail. +20% daily transactions, +2.2 NPS. Use for: multi-surface mandate, growth outcomes, leadership proof.
CS-02 Thermo Fisher. Exception-first regulated system with non-negotiable human release gate. Always attribute to BCG Digital Ventures. Use for: regulated environments, human-in-the-loop gates, enterprise complexity.
CS-03 American Red Cross. Multi-role shared records system. The design challenge is authorization boundaries that protect sensitive information while enabling cross-role coordination on shared objects. Routes differently by target company — shared records and multi-role coordination for platforms like Gusto, auditable authorization for compliance contexts like Stripe, escalation handoffs for operations-heavy environments, identity verification for companies like Checkr. If two companies get identical match language, the match is wrong. (Carried forward from prior coverage; not re-verified against the live page this cycle.)
CS-04 Allē. 3.2x redemption, 47% lapsed-member reactivation. Use for: growth outcomes, activation, retention, re-engagement.
CS-05 Equinox+. Connected fitness and digital membership experience. Translating a premium physical brand into a subscription digital product with engagement, retention, and workout-experience design. Use for: consumer product craft, engagement design, subscription experience, brand-to-digital translation. (Carried forward from prior coverage; not re-verified against the live page this cycle.)
CS-06 Carrier IQ. Eight carrier portals, parallel extraction, completeness-based confidence, side-by-side broker comparison. Richest current control surface. Evidence boundary: the application shows evidence → review → note → verification → approval. It does not show disputed output → accepted correction → changed system → evaluated rollout → verified later behavior. Do not claim the correction-to-next-run chain is viewable.
Trust Essay and Five Handoffs
Trust Is the New Interface — five handoffs:
- Intent-Setting. Structured intake, confirmed interpretation, success defined before execution.
- In-Progress. Visible steps, sources, accumulating work.
- Output Review. Evidence, provenance, confidence, what the system did not find.
- Decision Gate. Enough context for a real decision; human retained where consequences are irreversible.
- Loop Feedback. One run's output entering the next; small errors compound without a feedback path.
Includes the Watch → Verify → Delegate ladder. The two-axis correction (autonomy scope x intervention sensitivity) was published in Portfolio Playbook Issue #7 but does not appear as a live portfolio artifact.
Per-Company Positioning
Two active companies this cycle. No Watch-tier companies carry forward.
Render — Staff Product Designer, Agent Experience
Posting: Live on Ashby, August 7. Remote US. $218K–$300K + equity. No named hiring manager.
Their problem: A coherent experience model across agent deployment, human review, agent error/recovery, and human access grants — spanning dashboard, CLI, API, SDK, MCP, Blueprints, Skills, and agent-facing documentation. Agents and humans share platform objects. Scoping, consent, token management, and revocation must work across both.
Lead with: The reversibility frontier (label it conceptual). Render's loop — deploy, review, error, recovery, access grant, revocation — is a sequence of diminishing recoverable choices. The core question: at what point can the agent's action no longer be stopped or undone, and does the interface make that visible? Maps directly to Decision Gate from the Trust essay.
Support with: Carrier IQ for staged execution with evidence, review states, and an approval boundary. Alibaba for multi-surface platform coherence.
Differentiator: A published trust framework with five named handoffs, a live review-oriented agent application, and platform-scale production case studies. Few candidates at this level have all three.
Gap: The portfolio contains no CLI, API, SDK, MCP, or agent-documentation case study. No distributed-cancellation or scoped-token proof. TinyFish may supply verbal context but cannot serve as the case study.
Hardest question they will ask: "Show me how you maintained coherence between a GUI dashboard and a CLI/API for the same operation." Alibaba's multi-surface mandate is the closest answer, but it does not include a non-GUI artifact. Flag as artifact opportunity.
OpenAI — Product Design Leadership, Growth
Posting: Live on openai.com, August 18. San Francisco. $347K–$405K + equity. No named hiring manager.
Their problem: B2B growth across Codex and ChatGPT for Business. Discovery → onboarding → activation → engagement → collaboration → conversion → retention. Player-coach managing 2–5+ designers while staying hands-on. Open to experienced managers and exceptional senior ICs. Experimentation, quantitative data, research, and qualitative evidence are named methods.
Lead with: Agentic Labs as current AI craft proof — three solo-built, live AI applications that demonstrate the hands-on execution judgment a player-coach role demands. The Trust essay demonstrates systematic thinking about AI interaction design that can inform team-level standards.
Support with: Alibaba for leadership and growth outcomes (+20% daily transactions, +2.2 NPS). Allē for activation and retention (3.2x redemption, 47% lapsed-member reactivation).
Differentiator: Platform-scale growth outcomes from prior leadership roles combined with current, inspectable AI applications. Most candidates at this level have one or the other.
Gap: No published AI-native B2B onboarding, activation, conversion, or retention loop. No published controlled growth experiment in an AI product context. The growth outcomes come from pre-AI product work.
Cluster context: OpenAI currently has five Product Design postings: Growth, Identity (Aug 14), Design Systems (Aug 14), Engineering Acceleration (Aug 6), and Payments (Aug 6). Five simultaneous design seats suggest a hiring wave. Identity covers human-to-agent and agent-to-agent authorization — overlaps with Decision Gate. Engineering Acceleration owns rollout/rollback and the observe → investigate → evaluate → decide loop — overlaps with the three-claim framework. These adjacencies strengthen the case that Juno's framework thinking fits OpenAI's current organizational investment.
Language rule: Do not use "design-led" language. Match their operating model: experimentation, data, cross-functional growth.
Positioning Rules and Attribution Requirements
These rules govern all generated outreach, cover letters, and interview prep. Some repeat constraints from other sections. The redundancy is intentional — this section must be extractable and applied independently.
- TinyFish is context, never proof. Past tense only. No screens, traces, links, or internal artifacts. The pricing essay is publicly accessible but not authorized.
- Thermo Fisher requires BCG Digital Ventures attribution in every reference: "work completed as Product Design Director at BCG Digital Ventures."
- UAT Sentinel is not authorized for any external communication. Do not reference, link, or describe.
- Conceptual artifacts must be labeled conceptual. The three-claim framework, reversibility frontier, and Inference-Aware UX are not published or shipped. Present them as thinking, not portfolio proof.
- Portfolio outcomes are Juno-published figures. Alibaba (+20% daily transactions, +2.2 NPS) and Allē (3.2x redemption, 47% reactivation) have not been independently validated. Do not present them as third-party verified.
- Match the buyer's operating model in language. Do not use "design-led" when addressing organizations where design reports into product leadership.
- Route the same case study differently by company. Red Cross leads for shared-object authorization at one company, escalation handoffs at another, identity verification at a third. If two companies receive identical match language, the match is wrong.
- Do not fabricate, overstate, or claim familiarity with a company's product that Juno does not actually have.
Anticipated Interview Questions
"How would you design a task that disconnects, requests input, fails, or receives a cancellation that may not be honored?"
Grounded in MCP Tasks formalizing durable task handles, reconnectable state, and cooperative cancellation.
Answer from Carrier IQ's staged execution, review states, operator notes, and approval boundary. Decision Gate supplies consequence-awareness logic.
Gap: no reconnect, input_required handling, cancellation acknowledgement, or reversibility-frontier view in the portfolio.
"How do you separate authorization of an outcome from authorization of the routes, credentials, and side effects used to reach it?" Grounded in AISI's report of 19 unsanctioned actions across intentionally permissive cyber-evaluation runs. Answer from Carrier IQ's structured intake and approval boundary. TinyFish as bounded verbal context only. Gap: no runtime scope-diff, denied-route record, or changed-post-denial behavior in the portfolio.
"When should a product expose cost, speed, or effort level to the user, and when should it choose automatically?" Grounded in provider controls now exposing materially different reasoning/speed/price tiers (GPT-5.6, Gemini 3.7 Flash, Opus 5). Answer from Decision Gate's consequence and reversibility logic. TinyFish may ground verbal trade-off discussion. Gap: Inference-Aware UX has no published artifacts. This is the domain's first and most likely interview question.
"Show how a human correction changed the system and how you verified later behavior improved." Grounded in Presence and Vercel describing proposed-change → evaluation → human approval → controlled rollout. Answer from Carrier IQ (evidence, notes, re-verification, approval) and Thermo Fisher (BCG DV) for a human QA gate. Gap: the three-claim framework and complete correction lineage remain conceptual. Highest-priority build. (See Six Artifacts Ranked by What They'd Prove.)
"How do you stay hands-on while improving a team's judgment and measurable outcomes?" Answer from Alibaba for leadership and outcomes at scale. Agentic Labs as current hands-on AI craft. Allē for growth-specific outcomes. Gap: no published AI-native B2B growth experiment or team-level design-quality mechanism.
"How do you maintain coherence across human and agent interfaces?" Answer from Alibaba's multi-surface enterprise redesign and Carrier IQ's review surface. Gap: no non-GUI or agent-documentation proof in the portfolio.
Prohibited Claims
- The Carrier IQ correction-to-next-run chain is not viewable, published, or demo-ready.
- TinyFish work is not inspectable, linkable, or available as a case study.
- UAT Sentinel: do not reference in any external communication.
- The pricing essay (what-do-you-count.html): do not reference or link.
- Conceptual artifacts are not published or shipped.
- Do not use "design-led" when addressing organizations where design reports into product leadership.
- Portfolio outcomes (Alibaba, Allē) are Juno-published figures, not independently validated.
- Do not describe Juno as "first designer" at TinyFish without shifting the proof burden to practice-definition evidence (Agentic Labs answers this).
- Do not fabricate, overstate experience, or claim familiarity with a company's product that Juno does not have.
- OpenAI Identity role overlap: The Product Designer, Identity posting covers human-to-agent and agent-to-agent authorization, permissions, and administration — a close enough match to Decision Gate that it warrants its own positioning block if Juno considers applying to multiple OpenAI seats.
- MCP Apps as intervention surface: The MCP Apps extension lets servers return sandboxed interactive HTML inside a host conversation, which could supply the production-standard mechanism for the monitoring, approval, and multi-step review surfaces that the reversibility frontier concept needs.
- AISI's authorization-path distinction: The AISI incident report separates authorized outcomes from authorized means (routes, credentials, identities, networks), and AISI's planned response includes finer network controls and purpose-built real-time monitoring — language worth tracking because it may appear in Render's or OpenAI's interview vocabulary.
- Carrier IQ re-verification timing: The correction-to-next-run record has been the highest-ranked unbuilt artifact for two consecutive cycles, and both Presence and Vercel have now published production analogues that make the gap more visible to any interviewer who has read either account.