What you're entitled to claim
You can reason about where humans set intent, inspect agent work, evaluate evidence, retain approval authority, and build confidence gradually. You have published conceptual architecture, live applications, and past production context that prove it.
Interaction judgment — the design reasoning that determines how AI products earn user trust through visible handoffs, structured decision points, and incremental delegation — is your strongest provable construct. Five roles on your current target list screen for it as a first-order filter. This dossier establishes what each evidence pillar lets you say, where entitlement stops, and which companies to deploy it against.
Production-code authorship and code-assisted prototyping are separate constructs with separate evidence. They get their own dossiers. This piece covers interaction judgment only.
The Trust essay
"Trust Is the New Interface" is publicly accessible and linked from your homepage. It provides conceptual architecture: five handoffs in an agentic workflow (intent-setting, in-progress visibility, output review, decision gate, loop feedback) and a trust ladder (Watch, Verify, Delegate) describing how users move from observation to reliance as evidence accumulates. It separates model accuracy from designed consistency — the right distinction for products where the AI will sometimes be wrong and the interface has to make that survivable.
What it proves. You have a coherent, published model for how trust forms between humans and AI systems. An evaluator can read it before they ever see an artifact.
Where it stops. The essay is a vocabulary layer. It cannot substitute for inspectable states, gates, and failure modes. It does not prove the framework has been implemented or validated through production outcomes. An evaluator screening for craft will read the essay as conceptual unless they see an artifact alongside it.
How to use it. Link it in outreach. Pair it with a Lab, always. The essay gives the Lab a framework to be read against; the Lab gives the essay evidence that it works in practice.
The Agentic Labs
Three live applications at junochen.com demonstrate different facets of interaction judgment. They prove different things. Know which one matches which conversation.
Brand Pulse proves you design interpretive oversight for multi-agent sensemaking. Editable mission definition, parallel source agents with visible states, source-labeled excerpts attached to synthesis, history/cancel/rerun/restore controls. A reviewer can see that you separate what the agents found from what the system concluded and that the user can inspect the seam. It does not show an approval gate, a correction control that alters future behavior, or a confidence field. You cannot claim Brand Pulse demonstrates human authorization over consequential action.
Retail Velocity proves you design evidence-backed operational prioritization. Agent-collected material becomes ranked recommendations with confidence labels (high/medium/low), supporting evidence, source-page details, and an inspectable activity trail. Two constraints: the application names TinyFish infrastructure on its interface (Search, Fetch, and Agents), so you cannot claim authorship of the underlying agent stack. And there is no visible human authorization gate or correction flow.
The TinyFish branding is a deployment consideration, not a disqualifier. Your claim covers the design layer — how you structured evidence presentation, confidence communication, and the activity trail — not the infrastructure executing the agents. If an evaluator notices the dependency and asks, name it directly and redirect: the agent stack is TinyFish infrastructure; the interaction design decisions — what the user sees, how results are ranked and justified, how confidence is communicated — are yours. Do not lead with Retail Velocity when the evaluator's primary screen is technical authorship. Lead with it when the screen is information architecture for agent-generated output.
Carrier IQ proves you design the boundary between agent preparation and accountable human authorization. Staged execution (Session, Navigate, Fill, Extract, Verify), differentiated review states for exceptions, evidence and trust signals attached to results, re-verification capability, and an explicit Approve Bind gate that retains human authority over an irreversible action. It does not publish a complete sequence showing a wrong output corrected, a versioned system change, a comparable rerun, and regression monitoring. It does not publish adoption data or field outcomes.
Carrier IQ is your strongest single artifact for interaction judgment. It demonstrates the most complete handoff chain: agent prepares, surfaces evidence, routes exceptions to appropriate review states, and withholds commitment until a human authorizes it. When a role screens for trust architecture, lead with Carrier IQ.
Your homepage links the Trust essay but does not link any of the three Labs, and legacy case-page routes return 404. Fix the routing or include direct application URLs in every outreach.
The applications are reachable at their standalone URLs, but a reviewer who arrives at your homepage and clicks around will never find them.
TinyFish context, past tense only
Your prior work at TinyFish involved production agent traces, auditability, attribution, reversibility, and governance in enterprise deployment. This is not portfolio proof. It is production context that validates the interaction-judgment reasoning your essay and Labs express.
What it proves. Working with agent traces in production taught you where human checkpoints actually need to go — not where they look logical in a wireframe, but where silent agent failures create real downstream risk. Governance work showed you how trust degrades when auditability is absent: users lose confidence on a single unattributed error and don't recover. Reversibility in enterprise deployment is a different problem than reversibility in a prototype — the cost of an irreversible agent action in production is what makes the Approve Bind gate in Carrier IQ a design requirement. These are things conceptual work alone cannot teach.
Where it stops. TinyFish work is not public portfolio proof. It cannot be cited as evidence that you personally wrote production code. Reference it as context — "my previous work in enterprise agent governance" — if it comes up. Do not introduce it yourself.
What all three prove together
Each pillar has stated limitations. The essay is conceptual. The Labs lack correction-loop evidence. TinyFish can't be cited as portfolio proof. But the composite is stronger than any pillar alone: you developed a published model for how trust forms in agentic systems, built live applications that test different parts of that model against real interaction problems, and your design reasoning was shaped by production deployment where the consequences of getting these handoffs wrong were operational. That composite — conceptual architecture, inspectable artifacts, production-informed judgment — is what you are entitled to claim.
The wall on production code
This evidence does not prove you write production code. The three Labs demonstrate interaction judgment — how you decide what humans should see, control, and authorize. They do not demonstrate that you built the systems executing those decisions in a production codebase.
If you let an evaluator infer that the Labs prove engineering capability, you create a verification surface you cannot defend. When that claim collapses, it damages the claims that were actually true. As covered in Issue #8, visibility, permission, attribution, and technical dependency cannot be collapsed into a general claim that all Labs are wholly independent work. The boundary protects everything on the interaction-judgment side.
Current gap classification
Interaction judgment is a hybrid gap — part perception, part evidence.
Perception: the evidence exists but evaluators can't find it. Your homepage doesn't route to the Labs. The case pages 404. A reviewer who doesn't receive a direct link sees the essay but not the artifacts that ground it. Fixable this week.
Evidence: no current artifact shows the full temporal proof unit — wrong output, diagnosis, intervention, comparable retrial, improvement, and confirmation that adjacent behavior didn't regress. The Delegation Contract was specified to close this gap. As of September 19, it has not shipped. No other Build Queue item targets correction-loop evidence for interaction judgment. Until the Delegation Contract ships, the gap remains open.
Which companies screen for this first
Five current roles screen for interaction judgment as a first-order filter.
OpenAI — Product Designer, Identity. Identity is the trust layer for every product OpenAI ships — authorization, permissions, sensitive-action confirmation, recovery, governance. The team designs who and what gets to act, under what conditions, with what oversight. Lead with the Trust essay plus Carrier IQ. Your evidence fits the Identity role specifically. The Codex posting requires recent production-code shipping, which this evidence cannot serve.
Render — Staff Product Designer, Agent Experience. Render is building the infrastructure layer where developers deploy agents, which means every supervision, access-control, and recovery decision they make becomes the default for thousands of downstream applications. The full deploy/review/error/recovery/access loop, plus human-agent handoffs and supervision surfaces. Lead with Carrier IQ plus the Trust essay; Retail Velocity can support cross-surface pipeline visibility. The role also asks for comfort reading and prototyping near code, and prior CLI/API design at scale. Interaction judgment gets you into the conversation. It may not close it alone.
Handshake — Staff Product Designer, Student Experience. An agent acting on students' behalf in consequential career decisions — job applications, employer outreach — where a wrong action has real professional cost and the user population has limited experience evaluating AI recommendations. The posting names reasoning visibility, appropriate trust, uncertainty handling, and handoff between AI and human judgment. Lead with the Trust essay plus Carrier IQ; Brand Pulse can supplement the reasoning-visibility discussion. The role also screens for consumer craft, experimentation, and measurable career outcomes.
Slack — VP Product Design, Principal Architect. Slack is positioning itself as the AI operating system for work, and the interaction patterns this role defines will govern how millions of users experience agent collaboration for the first time. Human-agent collaboration, agent action, handback of control, dynamic and ambient interfaces. Lead with the Trust essay plus Carrier IQ. The role demands visionary concept work, rapid prototyping, enterprise scale, and emerging modalities (voice, ambient, generated interfaces). Your three pillars don't prove experience in those modalities.
Stripe — Senior Staff Product Designer, Data & AI. Stripe processes payments. Every AI-assisted decision in their product carries financial consequence, which makes trust architecture a business requirement. Systems thinking, ambiguous product direction, high-stakes trust. Lead with the Trust essay plus whichever Lab matches the conversation — Carrier IQ for decision authority, Retail Velocity for evidence-backed prioritization. The posting does not identify production-code authorship as a requirement, which works in your favor.
Response phrasing
Language for live conversations. Confidence markers in parentheses.
When asked about your approach to human-AI trust:
"I published a framework for this — five handoffs where trust forms or breaks in an agentic workflow, from intent-setting through loop feedback. Then I built three applications that test different parts of it. Carrier IQ is the clearest example: the agent does preparation and evidence gathering, but binding authority stays with the human until they've reviewed differentiated exception states." (High confidence. Fully supported by public evidence.)
When asked how you decide what the human should control:
"The question I start with is what happens if the system is wrong at this step. If the consequence is reversible, the agent can act and the human reviews after. If it's irreversible — like binding an insurance policy — the system prepares and recommends, but commitment requires an explicit human gate. Carrier IQ is built around that distinction." (High confidence.)
When asked about your AI experience broadly:
"My focus is interaction judgment — how humans and AI systems hand off control, how you surface uncertainty without creating paralysis, how you build trust incrementally. I've published on it and built working applications that demonstrate it. I'm not claiming to be a production engineer." (High confidence. The last sentence is optional but recommended if you sense the evaluator is probing for code. Volunteering the boundary early protects everything else.)
When asked about TinyFish:
"My previous work in enterprise agent governance gave me direct experience with the production constraints — auditability, attribution, reversibility — that shaped how I think about these handoffs. I can speak to the design reasoning; the product details are under NDA." (Moderate confidence. Use only if TinyFish comes up organically.)
When asked about correction loops or how the system learns from mistakes:
"That's the temporal dimension — how one run's output becomes the next run's input and how you prevent small errors from compounding. I address it in the essay's fifth handoff, loop feedback, and it's an area I'm actively building toward in the Labs. I don't have a shipped public artifact that demonstrates the full correction cycle yet." (Use with caution. Honest and protects credibility, but surfaces a gap. Deploy only if the evaluator asks directly.)
Quick reference for pre-interview scanning
Your claim: You design how humans and AI systems hand off control, surface uncertainty, and build trust incrementally. You have a published framework, three live applications, and past production context that together demonstrate interaction judgment across conceptual architecture, inspectable artifacts, and production-informed reasoning.
Your wall: This evidence does not prove you write production code. Say so if asked.
Lead artifacts by company:
| Company | Lead with | Support with |
|---|---|---|
| OpenAI Identity | Trust essay + Carrier IQ | — |
| Render | Carrier IQ + Trust essay | Retail Velocity |
| Handshake | Trust essay + Carrier IQ | Brand Pulse |
| Slack | Trust essay + Carrier IQ | Brand Pulse |
| Stripe Data & AI | Trust essay + best-fit Lab | — |
Before the conversation:
- Confirm the three Lab URLs are reachable. The homepage won't get evaluators there.
- Have the Trust essay URL ready to drop in chat or follow-up email.
- Know which Lab matches this company's problem. Carrier IQ for authorization roles. Retail Velocity for prioritization and evidence pipelines. Brand Pulse for multi-agent interpretive oversight.
The gap you'll be asked about: The full correction cycle — wrong output to diagnosis to fix to verified improvement. The Delegation Contract was designed to close this. It hasn't shipped. If asked, acknowledge it directly and describe what you're building.
- Render's agent experience scope: The Staff Product Designer, Agent Experience posting asks for comfort reading code and prior CLI/API design at scale, which means interaction judgment alone may not close this role without supplementary prototyping evidence.
- Stripe's nonlinear-career language: Stripe's Data & AI posting explicitly welcomes applicants whose careers contain unusual turns, which may reduce the category-spanning penalty that peer-reviewed research associates with nonlinear professional sequences.
- Docebo's hybrid screen: Docebo's Senior Product Designer, AI Experience role asks for both interaction-judgment constructs and code-assisted prototyping with Claude Code or Cursor, making it a useful test of whether the interaction-judgment pillar alone gets past the first gate when prototyping is co-screened.
- The Delegation Contract backlog: Portfolio Playbook's Issue #11 specified the Delegation Contract as one artifact consolidating ten prior build items, and its status as of September 19 remains "specified, not built" — the single highest-value build for closing the correction-loop evidence gap.