Version: 5.0 Date: August 30, 2026 Status: Active
Trust essay reframe: supervisory posture replaces accumulated trust; control clauses replace handoffs. Two artifact registrations (paired argument, Brand Pulse extension). Capital One posting refresh confirmed August 21 — third iteration of this requisition.
Identity and Background
Name: Juno Chen Location: San Francisco Bay Area Site: Juno Chen — Design Leadership for High-Stakes Systems Positioning line: "I build design teams and ship products where the system is complex, the stakes are real, and the design has to hold."
Production record spans B2B platforms at enterprise scale (Alibaba.com, Head of Design & Research, North America), regulated and high-consequence systems (American Red Cross disaster relief, Thermo Fisher pharmaceutical supply chain), and consumer product launches (Equinox+, Alle). All production case studies except Alibaba.com were executed under her BCG Digital Ventures title. That attribution is mandatory in every reference.
TinyFish (past tense only). Most recent role. Shipped AI-native enterprise tools, wrote the Trust framework, built working agents. Establishes recency and technical credibility. Never cited as a case study. Never used as portfolio evidence. Never sourced for outcome claims.
Trust Essay — Control Clauses and Supervisory Posture
Published: "Trust Is the New Interface"
The v5 Correction
Prior versions described the trust ladder (Watch, Verify, Delegate) as accumulated trust in the agent and the five handoffs as journey stages. Both were wrong.
Trust ladder = supervisory posture, scoped to a specific delegation. A user may delegate one task to an agent while watching the same agent on a different task. The posture belongs to the task scope, not the agent.
Five control clauses = structural points where the user inspects, adjusts, or revokes. They are not sequential and they are not a journey.
| Control clause | What it governs | Delegation authority |
|---|---|---|
| Intent-Setting | Structured intake, confirmed interpretation, success criteria before execution | Establishes the bounded delegation: authorized scope, constraints, what the agent may and may not do |
| In-Progress | Visible steps, sources, accumulating work | User detects variance from authorized scope |
| Output Review | Evidence, provenance, confidence, what the system did not find | User evaluates result against the delegation's success criteria |
| Decision Gate | Context for consequential or irreversible decisions | User retains authority over outcomes the delegation did not authorize the agent to finalize |
| Loop Feedback | One run's output becomes input to the next | User changes terms of subsequent delegations based on what this run revealed |
The published essay already contains control-oriented material: Intent-Setting confirms interpretation, Decision Gate reserves irreversible decisions, the conclusion treats loop boundaries as gates. But the essay organizes these as "handoffs" and "transitions" through a workflow. It does not define delegation by task scope, permitted tools, cost ceiling, or irreversibility threshold. The v5 reframe extends beyond the essay's own framing.
Generation rule — vocabulary: Use control-clause language, not the essay's "handoff" vocabulary. Do not describe the ladder as accumulated trust or durable progression.
Generation rule — attribution: The published essay uses "handoffs," "transitions," and "trust ladder." The control-clause and bounded-delegation framing is Juno's current thinking, articulated in this DCF, not language that appears in the essay. When referencing the essay directly, say it establishes the five structural points, then extend to control-clause framing as Juno's evolved articulation. Do not write "as I argued in my Trust essay" followed by control-clause language the essay does not contain.
Forward-Looking Artifact Inventory
| Artifact | Domain | Status | Control-clause mapping | Primary company relevance | Lab connection |
|---|---|---|---|---|---|
| Paired argument artifact (delegation envelope + cancellation settlement) | Human-Agent System Design — Task State and Intervention | Conceptual. Not published. Not in progress on any live page. | Intent-Setting (bounded delegation), In-Progress (variance/cancellation state), Output Review (re-entry packet), Decision Gate (irreversible effects), Loop Feedback (revised terms) | OpenAI, Vanta, Headway, Capital One | Carrier IQ (Lab 04) is the closest published precursor: the broker retains the pricing decision while the agent assembles — a delegation boundary the paired argument artifact would formalize. |
| Brand Pulse extension (artifact inheritance map) | Agent Infrastructure as UX | Base Brand Pulse published. Inheritance extension conceptual and absent. | Output Review (provenance), Decision Gate (who/what may act from the artifact), Loop Feedback (correction, supersession, dependency invalidation) | Vanta, Capital One, OpenAI | Brand Pulse (Lab 01) is the direct base. |
Both artifacts were specified in prior editorial work. Neither has shipped. When referencing in outreach, name them as thinking-in-progress. Do not describe as shipped or in-progress work.
Agentic Labs — Published Foundations
The homepage presents four agent demonstrations under the heading "Three live agents built solo" (the count mismatch exists in the current server HTML — do not correct it in outreach). Three are approved for external use.
| Lab | Name | What it demonstrates | Control clauses active | Outreach status |
|---|---|---|---|---|
| Lab 04 | Carrier IQ | Parallel carrier-portal access, structured extraction, completeness-based confidence, side-by-side comparison. The broker evaluates while the agent assembles. | Output Review (provenance, confidence), Decision Gate (broker retains pricing decision) | Approved. Homepage says five carriers / 90 seconds; Lab 04 brief says eight portals / four minutes. Do not assert either figure. Use the design-problem framing. |
| Lab 01 | Brand Pulse | 76 live signals become one decision surface. The interface concentrates on what requires human judgment. | In-Progress (visible accumulation), Output Review (evidence density beyond manual review) | Approved. Base for the Brand Pulse extension artifact. |
| Lab 02 | Retail Velocity | One agent audits 247 venues. The interface surfaces what the human should act on. | Output Review (exception surfacing), Loop Feedback (audit-to-audit improvement) | Approved. |
| Lab 05 | UAT Sentinel | Seven-phase QA pipeline retaining engineering context while the agent runs. | In-Progress (phase visibility), Decision Gate (engineering sign-off) | Not classified. Do not reference externally. |
| Lab 03 | Creator Scout | YouTube/TikTok parallel search, creators ranked by engagement, audience fit, posting velocity. | Output Review (ranking transparency) | Not approved. Not linked from homepage. Do not reference. |
All Lab brief pages label themselves "/ 06" but no sixth Lab is published or identified.
Case Studies CS-01 through CS-06
All outcome figures are self-published. BCG Digital Ventures attribution applies to CS-02 through CS-05.
CS-01 — Alibaba.com
Head of Design & Research, North America. Desktop: 25% of traffic, 80% of transaction value, designed like a consumer marketplace. Juno named the procurement gap and secured the redesign mandate across homepage, search, product detail, onboarding, checkout. Reported: +7% DAU, +20% transactions, +2.2-point NPS, 47% reduction in reported security concerns.
CS-02 — Thermo Fisher / mySupply
Product Design Director, BCG Digital Ventures. Six pharmaceutical partners, nine sites, coordinated through spreadsheets and calls. 32 interviews, five failure modes became five modules with exception-first operational, forecast, capacity, and batch views. Reported: zero-to-live in 12 months, 100% partner adoption, $20M margin opportunity, 83% IRR.
CS-03 — American Red Cross
Product Design Director / GM, BCG Digital Ventures. Six disconnected systems, untrained surge volunteers, caseworkers, finance officers, supervisors under federal oversight. One record with role-specific views spanning intake, verification, disbursement, approval, event command. Reported: national deployment in six months, 1,689 cases in first two weeks, $847K disbursed, six systems to one.
CS-04 — Alle
Product Design Director, BCG Digital Ventures. Loyalty program for 30M members designed around product distribution rather than patient or provider practice. Redesigned consumer iOS and provider backoffice around one loyalty model. Reported: 3.2x redemption, 47% reactivation, CAC from $92 to $42. (Site labels this Showcase 05; DCF numbering follows v4 convention.)
CS-05 — Equinox+
Product Design Director, BCG Digital Ventures. Five fitness brands, 600K premium members, brand identity preservation required. Four designers and one researcher, product definition through zero-to-MVP in 90 days. Reported: 4.8-star App Store rating at launch. (Site labels this Showcase 04; DCF numbering follows v4 convention.)
CS-06 — Carrier IQ
Solo-built Lab exploration (see Lab 04). Working demonstration, not a client case. Use it to prove design thinking in agentic systems. Do not claim enterprise outcomes.
Cummins / ZED Connect
BCG Digital Ventures project (BCG DV attribution mandatory). Industrial IoT / connected-vehicle platform. Not one of the six canonical case studies. Insufficient verified detail for a full summary or outcome claims. If a company asks about supply-chain or industrial-IoT experience, mention briefly with BCG DV attribution as additional scope evidence alongside CS-02 (Thermo Fisher). Do not present as a case study or source outcome figures.
Per-Company Positioning
Act Tier
Vanta — Head of Design
Posting | Remote U.S. | ~40 reports | $365K–$513K + equity | Active August 30.
The posting asks for agentic and AI-native product experiences, AI-native design practice, redesigned team workflows around AI tooling, design systems, definitions of done, launch-review frameworks. Requires org-level AI fluency and B2B SaaS depth.
Lead with: Paired argument artifact (conceptual) mapped to Vanta's risk-object and agentic-workflow language. The delegation envelope addresses how a compliance team delegates monitoring to an agent and knows when to intervene. Bridge to Trust essay control clauses as the design framework underneath. Support with: CS-02 (Thermo Fisher) for regulated multi-stakeholder complexity. CS-03 (Red Cross) for operational systems under institutional oversight. Ask about: How the current design org handles the boundary between agentic automation and human review in GRC workflows. This surfaces whether the role is building that boundary or inheriting one.
Headway — Staff Product Designer, Provider CRM & Prospect Management
Posting | $212K–$265K + equity | Active as of research cutoff.
Three-party delegation problem. Provider, agent, patient. Per-action review is structurally impossible — the patient cannot inspect each action the agent takes on the provider's behalf. The design problem is where to draw trust boundaries around AI acting for providers when patients are downstream.
Lead with: Trust essay control clauses, specifically Decision Gate: who authorizes what the agent does when the patient cannot review each action. Carrier IQ as closest published proof of designing evaluation interfaces for agent-assembled outputs. Support with: CS-03 (Red Cross) for systems where undertrained operators make consequential decisions under time pressure. Ask about: How the current CRM surfaces provider actions that affect patients downstream, and whether there is an existing review layer or whether the role designs one from scratch. Different problem, different timeline, different organizational support required.
Watch Tier
OpenAI — Product Design Lead, Growth — Codex
Posting | Active as of research cutoff | $347K–$405K + equity.
Capability-adoption mandate: designing how an expanding system's capabilities reach users.
Lead with: Paired argument artifact (conceptual) and Brand Pulse extension — both address how users set intent and evaluate agent-produced outputs. Support with: CS-01 (Alibaba.com) for growth-surface redesign at scale. CS-04 (Alle) for redesigning engagement with a platform whose capabilities outpace the user's mental model.
Capital One — Director, Product Design — AI in Experience Design
Posting | McLean / SF / NYC | Posted August 21, 2026 | R249772 | Active August 30.
Organizational-capability mandate. Third iteration of this requisition. Prior analysis documented the May to July to August language shift: visionary CDO peer to operational specificity to enablement infrastructure. The posting now names Claude, Gemini, Claude Code, Stately/XState, and Windsurf explicitly. It asks for scaling AI processes and tools in a design organization and balancing risk with rapid innovation.
Three months unfilled. The committee has calibrated through two failed slates, the pipeline is thin at reset, and early candidates matching the revised language get outsized attention. The first two weeks after a reset are the highest-value window.
Lead with: Trust essay as installable methodology for evaluating when AI-assisted design work earns the right to ship. Control-clause framework maps to their need for governance structure around AI tooling adoption. Support with: CS-02 (Thermo Fisher) for operational infrastructure across distributed teams. Brand Pulse extension (conceptual) for artifact provenance and inheritance in multi-agent contexts. Mirror their tool vocabulary. The posting names specific tools. If Juno has direct experience with any of them, say so in the first paragraph of outreach.
Positioning Rules
- TinyFish is past tense. Recency and AI-native credibility only. Never a case study, never portfolio evidence, never a source of outcome claims.
- BCG Digital Ventures attribution is mandatory on Thermo Fisher, Red Cross, Equinox+, Alle, and any Cummins/ZED Connect reference.
- AI-native targets: frontier artifacts first, production case studies second. Enterprise and regulated targets: production case studies first, Trust essay as methodology second.
- Conceptual artifacts must be labeled conceptual. Paired argument and Brand Pulse extension are thinking-in-progress. Name them as frameworks Juno is developing.
- Carrier IQ scale/timing conflict is unresolved. Do not assert "five carriers in 90 seconds" or "eight portals in four minutes." Use the design-problem framing: parallel access, structured extraction, completeness-based confidence.
- UAT Sentinel and Creator Scout are not approved for outreach.
- Trust essay language: control clauses, not handoffs. Describe the five structural points as control clauses governing a bounded delegation. Do not use "handoff," "journey stage," "accumulated trust," or "durable progression."
- Do not use DCF or site numbering in generated outreach. The DCF and the live site number case studies differently. Use project names (Alibaba.com, Thermo Fisher, Red Cross, Alle, Equinox+, Carrier IQ) without CS- or Showcase numbers in any candidate-facing material.
- anchor.is URLs are internal DCF references. They cite Portfolio Playbook editorial analysis, not Juno-authored content. Never attribute them to Juno. Never include them in generated outreach, cover letters, or candidate-facing material.
Anticipated Interview Questions
"How do you think about trust in AI systems?" Start with the control-clause framework: trust is a supervisory posture the user adopts for a specific task scope, not a property of the agent. Five control clauses — Intent-Setting, In-Progress, Output Review, Decision Gate, Loop Feedback — are structural points where the user inspects, adjusts, or revokes. Carrier IQ demonstrates Output Review and Decision Gate in production. Brand Pulse demonstrates In-Progress and Output Review. Reference the published essay, then extend to bounded-delegation framing as current thinking.
"Show me a case where design drove business outcomes." Alle (BCG DV): 3.2x redemption, 47% reactivation, CAC from $92 to $42. The design move was restructuring the loyalty model around the patient and provider practice rather than product distribution. Alibaba.com: +20% transactions by redesigning the procurement experience for the surface generating 80% of transaction value.
"How do you build and scale a design team?" Red Cross (BCG DV): built a system that untrained surge volunteers could operate under federal oversight — a design-team scaling problem expressed as a product problem. Equinox+ (BCG DV): four designers and one researcher, zero-to-MVP in 90 days across five brands.
"What's your experience with AI tooling in design workflows?" TinyFish context (past tense): shipped AI-native enterprise tools in production. Trust essay as the framework for evaluating when AI-assisted work earns the right to advance. For Capital One specifically, mirror their tool vocabulary if Juno has direct experience with the named tools.
"What happens when an agent fails or produces a bad result?" The control-clause framework defines where failure becomes visible (Output Review) and where the system should change (Loop Feedback). Carrier IQ's completeness-based confidence is a working example of surfacing uncertainty before the user acts. The paired argument artifact — currently conceptual — addresses what happens after cancellation or failure: which effects are reversible, which are not, who owns the residual state, and how the next delegation changes. Name it as active thinking, not shipped work.
Prohibited Content
- TinyFish as a case study, portfolio piece, or source of outcome claims
- Creator Scout or UAT Sentinel in any external material
- Carrier IQ scale/timing figures without resolving the homepage/brief discrepancy
- Trust ladder described as accumulated trust, durable progression, or agent property
- Five handoffs described as journey stages or sequential transitions
- Control-clause or bounded-delegation language attributed to the published essay's text
- Unattributed BCG DV project references
- Outcome figures presented as independently validated (all are self-published)
- Cummins/ZED Connect as a full case study or source of outcome claims
- Paired argument artifact or Brand Pulse extension described as shipped or in-progress work
- anchor.is URLs in any candidate-facing material
- DCF or site case-study numbering in generated outreach
- MCP cancellation is cooperative: The MCP Tasks specification explicitly warns that a cancellation request does not guarantee the underlying operation stopped, which means any interview answer about agent failure needs to distinguish "cancel requested" from "effects contained."
- Vanta's title shifted: The live posting now reads Head of Design, not "Head of Product Design" as previously captured — a small change that may signal scope or reporting-line adjustments worth monitoring across the next repost cycle.
- Double-blind evaluation precedent: DeepMind and MLCommons piloted a cryptographically protected AI evaluation where neither party could see the other's sensitive asset, establishing a separation-of-duties model that maps directly to Decision Gate and Output Review if Juno needs to discuss evaluation integrity in interviews.
- Capital One's third requisition iteration: The August 21 repost of R249772 now names specific tools (Claude, Gemini, Claude Code, Windsurf) — language that was absent in earlier iterations, confirming the committee's shift from visionary hire to operational enablement.