Version: 6 Date: September 4, 2026 Status: Canonical reference for AI-generated positioning, outreach, and interview materials v6 change: Adds attention contract dimension across all five handoffs. Updates artifact build status. Refreshes company set against September 2026 postings. Adds interview questions from GPT-6 Astra, Fable 5.1, MCP Tasks, and inference economics developments.
1. Identity and Positioning Rules
Last verified: September 4, 2026
- Name: Juno Chen
- Location: San Francisco Bay Area
- Canonical spelling: Carrier IQ (two words), matching site usage.
- Title preferences: Head of Design, Design Director, VP Design. Never generate "Head of Product" or "Product Director." This is a screening risk — it kills candidacy before a human reads anything else.
- Positioning line: "I build design teams and ship products where the system is complex, the stakes are real, and the design has to hold."
- BCG DV attribution rule: Every reference to Thermo Fisher, American Red Cross, Equinox+, Allē, or Cummins/ZED Connect must include: "work completed as Product Design Director at BCG Digital Ventures." Alibaba and Agentic Labs sit outside this rule.
- TinyFish rule: Past tense only. Credibility signal for AI-native production experience, not portfolio proof. No outcome claims, screens, metrics, or architecture details.
- Numbering rule: Use project names in candidate-facing material. The live site route order and in-page case labels currently conflict.
2. TinyFish Context
Last verified: September 4, 2026
- Role: Product lead at TinyFish, an enterprise web-agent platform
- Domain: Agentic platforms, human-in-the-loop, trust and oversight, AI-native product
- What she built: Shipped an enterprise web-agent platform from zero to production. Worked with traces, auditability, attribution, reversibility, and governance in a production AI system.
- What she learned: How AI systems actually behave in production — where they fail, what users need to trust them, how oversight works when agents act rather than suggest.
- Bridge narrative: "I went to product to learn how AI actually works. I came back to design because I missed making things people can feel."
- Homepage status: TinyFish appears in Selected Work as "Shipped 2025–2026," tagged NDA-protected. The card is visible but the underlying work is not inspectable.
- Why she left: Nothing verified. Do not generate departure reasons.
3. Forward-Looking Artifact Inventory
Last verified: September 4, 2026
Six cycles, zero forward-looking artifacts published. Every item below is specified but unshipped. Do not present any artifact as published or live in generated materials.
Attention contract specifications here inherit from Section 6, which governs. If the two sections diverge, Section 6 wins.
| Artifact | Domain | Handoff mapping | Attention contract spec | Build status | Serves |
|---|---|---|---|---|---|
| Delegation envelope | D2 Intent-Based Interaction, D4 Human-Agent System Design | Intent-Setting, Decision Gate | Defines what attention the user commits at task launch. Specifies which actions require re-engagement vs. proceed autonomously. Watch: inspects the complete envelope. Verify: accepts system-proposed defaults with review. Delegate: authorizes a standing envelope with expiry and reauthorization conditions. | Specified. Not published. | Named: OpenAI Identity, Altana. Pattern: any authorization-first role. |
| Cancellation settlement | D4 Human-Agent System Design | In-Progress, Decision Gate, Loop Feedback | Surfaces surviving effects that require user attention after cancellation. Maps obligation owners and discharge conditions. Watch: reviews each surviving effect. Verify: receives summary with drill-down on unresolved obligations. Delegate: receives only items requiring human judgment to close. | Specified. Not published. | Named: none currently assigned. Pattern: any role involving irreversible agent action. |
| Contract-variance disclosure | D1 Inference-Aware UX, D3 Agent Infrastructure as UX | In-Progress, Output Review | Alerts user when routing changes alter capability, cost, latency, or data handling. Watch: sees all routing changes. Verify: sees material changes only. Delegate: sees nothing unless the change exceeds the original delegation boundary. | Specified. Not published. | Named: none currently assigned. Pattern: roles where model switching is a product concern. |
| Confidence/authority matrix | D2 Intent-Based Interaction | Decision Gate | Shows that high confidence does not imply broad authority. Attention stays mandatory for unauthorized action regardless of model performance. Supervisory posture changes evidence depth, not the authority requirement. | Specified. Not published. | Named: OpenAI Identity, Altana. Pattern: any role where agent autonomy and human authorization coexist. |
| Correction lineage | D3 Agent Infrastructure as UX | Loop Feedback | Three-claim record: receipt, incorporation, verified improvement. Watch: reviews each correction. Verify: samples governed changes. Delegate: monitoring runs until a defined threshold or regression triggers re-entry. | Specified. Not published. | Named: none currently assigned. Pattern: regulated and compliance roles. |
4. Agentic Labs Foundations
Last verified: September 4, 2026
Brand Pulse — case page | live app
- Handoff demonstrated: In-Progress (continuous monitoring, hourly score), Output Review (evidence-backed signals, source attribution, Agent Trace)
- Extends toward: Contract-variance disclosure — already surfaces sources and classification; the extension would surface when the underlying model or routing changes
- Visual foundation: Brand Score, Competitive Snapshot, Sentiment & Sources, Theme Breakdown, Agent Trace, rerun and cancel controls
Retail Velocity — case page | live app
- Handoff demonstrated: Output Review (0–100 scoring, ranked placement gaps, confidence), Decision Gate (opportunity scores directing field action)
- Extends toward: Delegation envelope — scores and ranks but does not yet let the user specify which actions the agent may take autonomously vs. which require approval
- Visual foundation: Opportunity, Confidence, Evidence, Agent Log surfaces. Visibly powered by TinyFish infrastructure.
Carrier IQ — case page | live app
- Handoff demonstrated: All five. Intent-Setting (profile, coverage, carrier selection). In-Progress (Session, Navigate, Fill, Extract, Verify stages). Output Review (Coverage Detail, Carrier Trust Signals, Carrier Verification). Decision Gate (Approve Bind with review states). Loop Feedback (Operator Note, rerun verification).
- Extends toward: Cancellation settlement (per-carrier state survives cancellation), confidence/authority matrix (quote evidence is separate from bind authorization)
- Visual foundation: Staged execution, Coverage Deltas, Agent Trace, disabled/retained decision gates, side-by-side broker comparison with completeness-based confidence
5. Case Studies
Last verified: September 4, 2026 Template note: Project name is the stable identifier. Route and in-page numbers are listed for reference only — do not use in candidate-facing material.
Alibaba.com | Route: cs-01-alibaba-teaser
- Type: Production case study
- Role: Head of Design & Research, North America
- Attribution: Direct hire. Not BCG DV.
- Scale: $50B+ GMV, 200K+ suppliers, 25M+ desktop sessions
- Outcomes: +7% DAU, +20% transactions, +2.2-point NPS, 47% fewer reported security concerns. Team grew from 6 to 12+.
- Tags: B2B marketplace, procurement UX, platform redesign, growth, enterprise trust
- Best for: Enterprise platform roles, mandate-building proof, team-scaling evidence
Thermo Fisher mySupply | Route: cs-02-supply-chain-teaser
- Type: Production case study
- Role: Product Design Director, BCG Digital Ventures
- Scale: Six partners, nine sites, 12 months
- Outcomes: $20M margin opportunity, 100% partner adoption, 83% IRR
- Tags: Pharma supply chain, data visualization, regulated enterprise, exception-first operations
- Best for: Regulated/compliance roles, 0-to-1 builds in constrained environments
Allē | Route: cs-03-alle-teaser (page displays "Case Study 05")
- Type: Production case study
- Role: Product Design Director, BCG Digital Ventures
- Scale: 30M+ members, 40K+ practices
- Outcomes: 3.2× redemption, 47% reactivation, $42 CAC vs. $92 industry average
- Tags: Consumer app, provider backoffice, loyalty, two-sided healthcare-adjacent growth
- Best for: Growth roles, consumer + provider dual-audience proof
American Red Cross | Route: cs-04-red-cross-teaser (page displays "Case Study 03")
- Type: Production case study
- Role: Product Design Director / GM, BCG Digital Ventures
- Scale: Six systems consolidated into one, national deployment in six months
- Outcomes: 1,689 cases in first two weeks, $847K disbursed
- Tags: Case management, financial distribution, regulatory compliance, mission-critical operations
- Best for: Compliance/regulated roles, 0-to-1 under institutional constraint
Equinox+ | Route: cs-05-equinox-teaser
- Type: Production case study
- Role: Product Design Director, BCG Digital Ventures
- Scale: 600K+ premium members, 200+ clubs
- Outcomes: 0-to-1 in 90 days, 4.8-star App Store rating at launch
- Tags: Consumer technology, iOS/Android, subscription engagement
- Best for: Consumer product roles, speed-to-ship proof
Carrier IQ | Lab 04 (no numbered case route)
- Type: Lab demonstration
- Role: Solo-built Lab exploration
- Attribution: Not BCG DV. Juno's own work.
- Scale: Working demonstration with eight carrier portals, four-minute quote return
- Outcomes: No customer-adoption or business-outcome metrics. Demonstrates design judgment in agentic systems.
- Tags: Agentic insurance, regulated review, provenance, confidence, human approval
- Best for: Any role where agent oversight, staged execution, or consequential-action design is the mandate
6. Trust Essay, Five Handoffs, and Attention Contract
Last verified: September 4, 2026 Source: Trust Is the New Interface Core argument: Accuracy is what the model achieves. Consistency is what the design delivers. The trust ladder (Watch, Verify, Delegate) describes supervisory postures for bounded delegation, not accumulated trust. This section governs attention contract specifications. Section 3 inherits from here. If the two diverge, this section wins.
Intent-Setting
- Production: Carrier IQ profile capture, coverage selection, carrier selection (five of eight selected by default)
- Labs: Carrier IQ initial screen; Brand Pulse brand/competitor/source input
- Artifact: Delegation envelope (D2, D4)
- Attention contract: Requires focused attention to outcome, deadline, stakes, reversibility, budget ceiling, permitted actions, evidence requirements, maximum commitment. Watch: inspects the complete envelope. Verify: accepts system-proposed defaults with review. Delegate: authorizes a standing envelope with expiry and reauthorization conditions.
In-Progress
- Production: Carrier IQ staged execution (Session, Navigate, Fill, Extract, Verify); Brand Pulse hourly monitoring
- Labs: Both apps surface agent activity during execution
- Artifact: Contract-variance disclosure (D1, D3); cancellation settlement (D4)
- Attention contract: Attention should be releasable unless the user is needed. Surface what can change now, what becomes irreversible next, and what proof is still unavailable. Watch: continuous observation available. Verify: milestone sampling plus material-variance alerts. Delegate: bounded cue for re-entry only.
Output Review
- Production: Carrier IQ Coverage Detail, Carrier Trust Signals, Carrier Verification; Brand Pulse evidence-backed signals, Agent Trace
- Labs: Both apps provide provenance and source attribution
- Artifact: Correction lineage (D3); contract-variance disclosure (D1, D3)
- Attention contract: Concentrated review of provenance, uncertainty, scope variance, surviving effects, and unresolved evidence — without requiring the user to reconstruct the full trace. Watch: reviews full evidence. Verify: receives summaries with drill-down. Delegate: receives exceptions and a re-entry packet when judgment is required.
Decision Gate
- Production: Carrier IQ Approve Bind with review states and disabled/retained gates
- Labs: Carrier IQ separates quote evidence from bind authorization
- Artifact: Delegation envelope (D2, D4); confidence/authority matrix (D2)
- Attention contract: Explicit attention stays mandatory where action is irreversible or outside the existing grant. Confidence cannot substitute for authority. Across Watch/Verify/Delegate: evidence depth may shrink with verified reliability, but the authorized human or organizational owner must still be established.
Loop Feedback
- Production: Carrier IQ Operator Note, rerun verification; Brand Pulse rerun and cancel controls
- Labs: Both apps allow re-execution; Carrier IQ preserves operator annotations
- Artifact: Correction lineage (D3); cancellation settlement (D4)
- Attention contract: Periodic organizational attention to correction scope, versioned change, evaluation threshold, release decision, and later comparable behavior. Watch: reviews each correction. Verify: samples governed changes. Delegate: monitoring runs until a defined threshold or regression triggers re-entry.
7. Per-Company Positioning
Last verified: September 4, 2026
No canonical September tier decisions exist. Last explicit assignments: DCF v5, August 30. Companies marked "unassigned" need an editorial tier decision before outreach goes out.
For unassigned companies: Positioning materials may be generated from the fields below but must be flagged as requiring tier confirmation before use in outreach.
OpenAI — Product Designer, Identity | Tier: Unassigned (requires reporting-line research)
- Lead: Carrier IQ (authorization, agent identity, provenance, confidence, human review map directly to the posting)
- Support: Alibaba (enterprise trust at scale), Trust essay (confidence/authority independence)
- Differentiator: Production experience designing systems where agents act with real consequences
- Evidence gate: Who does this role report to? Answer determines platform-authority role vs. surface-design role.
OpenAI — Product Design Lead, Growth, Codex | Tier: Last known Watch; engagement moved to Probe Now
- Lead: Allē (growth metrics, two-sided engagement, reactivation)
- Support: Alibaba (mandate-building, DAU/transaction growth), Trust essay for AI fluency
- Differentiator: Growth evidence with design-leadership scope
- Evidence gate: Is this still a growth-metrics role or has Codex matured past the growth-experimentation phase?
Altana | Tier: Unassigned
- Lead: Carrier IQ (multi-source verification, confidence scoring, staged agent execution)
- Support: Thermo Fisher (supply chain, regulated enterprise, partner adoption), Red Cross (mission-critical operations)
- Differentiator: Has designed for provenance and trust in systems where downstream consequences matter
- Evidence gate: Does design report to product, engineering, or an executive with design authority? The posting names authorization, consent, and interruption but says nothing about org structure.
Gusto | Tier: Process Check (August 29)
- Lead: Red Cross (compliance, financial distribution, multi-role records)
- Support: Thermo Fisher (regulated enterprise), Alibaba (platform scale)
- Differentiator: Has shipped financial-consequence systems under regulatory constraint
- Evidence gate: Is the posted role a backfill or a new seat?
Amplitude | Tier: Unassigned. Prior analysis found this Head role primarily tests organizational leadership, with AI-tool fluency as secondary.
- Lead: Alibaba (team scaling from 6 to 12+, org-building under platform complexity)
- Support: Equinox+ (speed-to-ship), Labs portfolio (AI fluency)
- Differentiator: Has built and scaled a design org inside a complex platform company
- Evidence gate: Does the Head of Design report to CEO, CPO, or CTO? That answer determines organizational authority vs. elevated IC.
Vocabulary sources (not tier-assigned): Ramp (coded prototyping language), Omada (consent, escalation, evaluation loops), Suno (consumer craft bar). Use for interview vocabulary calibration only. Do not generate outreach.
8. Anticipated Interview Questions
Last verified: September 4, 2026
Q: "What must the product prove after a task is paused or stopped?" Triggered by: GPT-6 Astra distinguishes interactive pause from API task termination; reports lower chain-of-thought monitorability than GPT-5.6. Answer: Lead with Carrier IQ. Staged execution means each carrier's state survives interruption. Per-carrier review states, Operator Notes, and disabled-but-retained decision gates demonstrate what persists after a task stops — not just that it stopped. Preemptive artifact (unbuilt): Cancellation settlement — surviving-effect inventory and obligation closeout record.
Q: "How do you show progress without pretending to expose reasoning?" Triggered by: Fable 5.1 surfaces user-readable progress messages while withholding internal reasoning. Answer: Brand Pulse shows sources and evidence without claiming to expose the model's internal process. Carrier IQ names execution stages (Navigate, Fill, Extract, Verify) and provides intervention surfaces at each. Progress is about what the user can act on, not about performing transparency. Preemptive artifact (unbuilt): Re-entry packet distinguishing progress evidence from hidden reasoning, with uncertainty markers and available intervention points.
Q: "How would you design cancellation when work or external effects may continue?" Triggered by: MCP Tasks specification defines cancellation as cooperative intent, not guaranteed termination. Answer: Carrier IQ retains per-carrier state after cancellation. Review states and Operator Notes persist. The Approve Bind gate can be disabled without erasing completed quotes. Cancellation is a settlement workflow, not a terminal state. Preemptive artifact (unbuilt): Cancellation settlement with phased states: requested, new authority revoked, in-flight identified, reversible compensated, residual assigned, closure verified.
Q: "Should users choose Fast, Standard, or Deep?" Triggered by: Astra Fast mode and Gemini effort controls expose speed, price, token use, and capability as selectable variables. Answer: Consequence and irreversibility should govern the design, not raw model capability. TinyFish production experience grounds the claim that inference economics shape product behavior. The user's interface should be stakes and constraints — deadline, spend ceiling, permitted actions, evidence depth — not model settings. Preemptive artifact (unbuilt): Commitment envelope translating model settings into user-legible terms: deadline, maximum spend, external calls permitted, evidence depth, irreversible exclusions.
Q: "How do confidence, expertise, and authority remain independent in your designs?" Triggered by: OpenAI Identity and Altana postings name authorization, agent identity, provenance, confidence, and human review as design territory. Answer: Carrier IQ separates quote evidence (confidence) from Approve Bind (authority). A high-confidence quote from a carrier does not authorize the bind. The confidence/authority argument is specified but not yet visualized. Preemptive artifact (unbuilt): Counterfactual showing high-confidence unauthorized action alongside low-confidence reversible action.
9. Prohibited
- Generating "Head of Product," "Product Director," or product-management titles for Juno
- TinyFish outcome claims, metrics, screens, architecture details, pricing details, or internal artifacts
- "First designer at TinyFish" unless supported by practice-definition evidence
- Expanding the bridge narrative into an account of why TinyFish employment ended
- Presenting any forward-looking artifact as published or shipped — all five remain specified and unbuilt
- Omitting BCG DV attribution from Thermo Fisher, Red Cross, Equinox+, Allē, or Cummins/ZED Connect references
- Generating case study content, metrics, or project details for Cummins/ZED Connect — no public case material exists; the attribution rule applies when the name appears but there is nothing to generate from
- Using CS numbers in candidate-facing material while site and DCF numbering conflict
- Generating outreach for vocabulary-source companies (Ramp, Omada, Suno)
- Generating outreach for unassigned companies without flagging that tier confirmation is required
- Treating secondary-market valuation estimates as equity evidence
- Claiming the attention contract, delegation envelope, commitment envelope, re-entry packet, or obligation map exists as published design work
- Fake quotes, fabricated company reactions, or invented interview dialogue
- Astra monitorability gap: OpenAI reports that GPT-6 Astra has lower chain-of-thought monitorability than GPT-5.6, which directly affects how much evidence a progress surface can actually draw from.
- Gusto posting age: The Gusto Head of Design listing has been live since July 22, making it over six weeks old at this cutoff — worth checking whether the funnel has closed or the role has been rewritten.
- A2A signed agent identity: The A2A Protocol v1.0 introduces signed Agent Cards and cross-organizational delegation, which maps to OpenAI Identity's agent-agent trust territory and could reshape the delegation envelope artifact.
- Enterprise safeguard burden: Anthropic's Enterprise Frontier Safeguards shift review to customer-designated humans, but neither the announcement nor independent coverage quantifies false-positive rates or reviewer workload — a gap that matters for any role designing operational review surfaces.