1. The Objection
"How deep is your AI experience, really — is it conceptual, or have you actually built?"
Every AI-native target on your list will ask some version of this during portfolio review or first-round screening. It sounds like one question, but it covers four.
2. What They're Actually Asking
The fear is that your AI work is something you have written about rather than something you have had to make decisions inside. An essay, some side projects, a product title at an AI startup: none of that proves you have sat with a model doing something unexpected and had to decide, on a deadline, what the interface should do about it.
That fear shows up as four separate screens depending on who is asking. Say "I've built three agentic systems" to an interviewer running a trust-and-oversight screen and you have answered the question. Say it to someone running a prototyping screen and they hear a non-answer.
The four constructs:
- AI interaction judgment — can you design trust, oversight, delegation, and human-agent handoffs for consequential decisions?
- Code-assisted prototyping — can you build working demos with Cursor, Claude Code, or equivalent tools?
- Production-code authorship — have you shipped code as part of your design role?
- Evaluation infrastructure — can you build prompts, graders, regression suites, model-release testing?
You are strong on the first. You can close the second. You have nothing for the last two, and those aren't gaps to fix — they are roles screening for an engineering-adjacent profile. Knowing which construct is active tells you whether to answer or to stop spending outreach time on that company.
3. The Strongest Case Against You
- Your Agentic Labs apps are live but invisible from the junochen.com homepage (verified September 19). Case-page routes return 404. An evaluator who checks your portfolio without a direct link sees no AI interaction work.
- No published artifact demonstrates code-assisted prototyping. No working demo, no Cursor-built flow, no prototype repository.
- TinyFish involved product leadership, not engineering commits. You have no production-code evidence.
- The Delegation Contract — the forward-looking artifact consolidating the four artifact domains (Inference-Aware UX, Intent-Based Interaction, Agent Infrastructure as UX, Human-Agent System Design) — remains unbuilt.
4. The Grain of Truth
Your AI interaction judgment is real: three live agentic systems, a published trust framework, daily production exposure to agent traces and governance at TinyFish. What fails is delivery. Someone landing on your homepage cold reaches none of it. The code-assisted prototyping gap is real in a different way — nothing published proves it. Production code and evaluation infrastructure aren't framing problems at all. You don't carry those capabilities.
5. Classification
Hybrid overall. But the classification splits by construct, and the split is the whole point:
| Construct | Classification | Driver |
|---|---|---|
| AI interaction judgment | Hybrid — strong evidence, broken delivery | Trust essay linked from homepage. Labs live but not navigable from homepage (verified September 19). Delegation Contract unbuilt. |
| Code-assisted prototyping | Evidence gap | No published prototype or repository. Verbal bridge only. |
| Production-code authorship | Fit boundary | Engineering capability requirement. Not a gap to close. |
| Evaluation infrastructure | Fit boundary | Anthropic's Evals role requires production Python, graders, regression suites. Specialized engineering function. |
Constructs 1 and 2 each warrant their own follow-up dossier with detailed evidence inventory, response phrasing, and target-company mapping. Constructs 3 and 4 are resolved here as fit boundaries — recognition and redirection, not preparation.
6. The Reframe (by Construct)
Construct 1 — AI interaction judgment. You have the evidence. The Trust essay supplies the conceptual architecture: five handoffs, Watch-Verify-Delegate. Carrier IQ shows it working in production through review states, an approval gate, and exception routing. Most recently as Head of Product at TinyFish, you shipped an enterprise agent platform from zero to one in three months, working daily with agent traces, auditability, and governance. What's broken is the route to all of it. An evaluator on junochen.com can't navigate to the Labs, and the case pages 404. Until routing is fixed and the Delegation Contract ships, send direct app links in outreach and walk the evaluator to the work yourself. High confidence in the reframe.
Brex's Staff Product Designer, AI screens for this construct specifically: delegation, supervision, correction, trust in automation touching customers' money.
Construct 2 — Code-assisted prototyping. Ramp's posting says it plainly: "comfortable with Cursor or Claude Code, even if you are not writing production code by hand." Brex asks for the same. This is becoming a baseline expectation at AI-native companies, and your portfolio has no published evidence of it. The verbal bridge: your Agentic Labs were solo builds that required working across design and implementation. That holds in conversation and does not survive a portfolio-only screen. Build Queue item: a published working prototype (Cursor or Claude Code output) demonstrating design exploration through code-assisted tooling. Status: not started. Moderate confidence in the verbal bridge live; low confidence without a published artifact.
Constructs 3 and 4 — Production code and evaluation infrastructure. When you detect either screen, recognize it and stop. Anthropic's Evals role requires production-quality Python, evaluation pipelines, and regression suites. Pursuing it with a verbal bridge wastes your time and theirs. Route that energy toward roles screening for constructs 1 and 2, where your position is strong or closable.
7. What to Say
Make the evaluator show you which screen you're on before you choose which evidence to spend.
"Can I ask what the AI design challenges look like day to day on this team — is it more about designing trust and oversight patterns for agent behavior, or more about prototyping directly in code with tools like Cursor?"
Then lead with the matching evidence.
If construct 1: "At TinyFish I shipped an enterprise agent platform from zero to one in three months — agent traces, auditability, governance, the full oversight architecture. I've published a trust framework for agentic handoffs and built three production systems that implement it. I can walk you through Carrier IQ's approval flow if that's useful."
If construct 2: "The Agentic Labs apps were solo builds — I've been working with AI-assisted tools throughout. I'm deepening my Cursor and Claude Code practice because that prototyping loop is where the design leverage concentrates right now."
8. What Not to Say
- "I've worked extensively in AI." Satisfies no specific construct. Signals you don't understand what they're screening for.
- "I published a trust framework." Against a prototyping screen, leading with the essay reads as conceptual. The essay is supporting evidence for construct 1 only.
- "I'm learning to code." If they're screening for production code, the honest position is that the role requires engineering depth you don't carry. Aspiration language confirms the gap without offering an alternative.
9. Early Signals
- Construct 1 is active when the interviewer uses trust, oversight, delegation, human-in-the-loop, consequential decisions, or asks about your philosophy on agent autonomy.
- Construct 2 or beyond is active when they mention specific tools (Cursor, Claude Code, v0), ask about your prototyping workflow, or use build to mean code-adjacent output rather than Figma files.
Quick-Reference Card
| Objection | "Is your AI work deep enough?" |
| Classification | Hybrid — four constructs, each classified separately |
| Real fear | That your AI experience is narrated (essays, frameworks) rather than operational (built systems, shipped code, hard decisions under model constraints) |
| Lead with | Diagnostic question to identify which construct is active, then Carrier IQ's approval flow (construct 1) or Agentic Labs solo builds (construct 2) |
| Say | "Can I ask what the AI design challenges look like day to day — trust and oversight patterns, or prototyping directly in code?" |
| Avoid | "I've worked extensively in AI" — generic framing that matches no specific construct |
Build Queue items:
- Fix Agentic Labs homepage routing and case-page 404s — status: outstanding
- Delegation Contract (consolidating four forward-looking artifact domains) — status: specified, not built
- Published code-assisted prototype — status: not started
- Brex combines two constructs: Their Staff Product Designer, AI posting explicitly screens for both trust-and-oversight judgment and code-assisted prototyping with Claude Code or Cursor — meaning construct 2 readiness directly affects your strongest construct 1 target.
- Ramp draws the line clearly: Their Product Designer posting explicitly states "comfortable with Cursor or Claude Code, even if you are not writing production code by hand," which is the cleanest current example of construct 2 separated from construct 3.
- Anthropic's fourth construct exists: The Product Designer, Evals & Prompts role requires production Python, graders, regression suites, and model-release testing — a specialized evaluation-infrastructure screen that goes beyond general production-code authorship.
- Atlassian separates IC from leadership evaluation: Their design interview handbook describes distinct assessment tracks where IC candidates face detailed craft questioning while management candidates face team-leadership questioning — useful for calibrating which evidence to foreground at companies running similar split processes.