Figma — Where Product Intent Lives When the System Is Legible to Agents

PHASE 1 — THE MOMENT

Trigger: Figma posted Director, Product Design — AI & Context on September 22. Eleven days old, full window value. Loredana Crisan's expanded scope from CDO to product-and-design leadership is the executive route in.

Real mandate: The posting names a "Context Platform Pillar" covering Design Systems, Code, MCP, CMS, and Community. Figma has stated publicly that three problems inside that pillar remain unsolved: where product intent lives once tokens and components become machine-readable, how drift between design and code gets governed when agents can write to both surfaces, and how human judgment travels alongside system rules. This Director owns the design architecture that resolves them.

Hiring cluster: Figma's careers page lists Product Designer roles for Design Systems, CMS, AI Models, and Roundtripping, plus a researcher for agentic experiences. Emily Campbell is in seat as Director of Design for AI Models. Figma is staffing AI-design leadership into several pillars on overlapping timelines, and that changes what the role is worth. Whoever lands first shapes team composition and how the pillars divide work. Whoever lands third inherits those decisions.

Urgency verdict: Act this week. Fresh posting, adjacent roles filling, buildout live. The Timeline Reconstruction traces Crisan's trajectory and the search urgency underneath it.

PHASE 2 — PORTFOLIO MAPPING

Positioning archetype: AI-native product design.

Figma's three open problems line up with the Trust essay's five handoffs, which is why the essay leads here instead of a case study. Each handoff is a moment where context has to pass between a human and an agent, and each one needs a different surface. Figma's skills — plain-language instructions that encode conventions and workflow behavior otherwise left tacit — are the intent-setting handoff running in production. Figma's framing of the canvas as the review surface, where generated work returns for inspection before going back to code, covers output review and the decision gate. Their own documentation separates what the MCP server sends (components, tokens, and the Code Connect mappings that tie a component to its implementation) from what the agent decides (which component to use, when to stop). The essay names that boundary and designs for it.

Lead evidence:

• Agentic Labs → CarrierIQ's provenance chain, from form completion to coverage selection to approval gate, shows the pattern at the artifact level. The approval gate, where a user inspects agent-produced context before authorizing a commitment, is structurally the same move as Figma's canvas-as-review-surface. Acknowledge the divergence when you use it: CarrierIQ authorizes a policy bind, Figma's canvas releases code. Technical evaluators notice when an analogy is overclaimed.

• TinyFish (technical currency, past tense): Most recently as Head of Product at TinyFish, you shipped an enterprise web agent platform 0→1 in three months, traced agent runs in production with LangSmith, and solved the cold-start problem — a blank input box replaced with progressive intent-setting. That gives you practitioner depth on the exact failure modes this role owns.

Competitive landscape:

• AI-native design leader (Anthropic/OpenAI pedigree): Strong AI brand recognition, unlikely to have published design thinking about context systems at this specificity, and unlikely to have design-systems depth. Your edge is that the Trust essay is a published context architecture with no obvious equivalent from this archetype. Confidence: moderate.
• Enterprise design-systems leader (Salesforce/Adobe): Systems depth and org scale, thin on AI-native credibility and on the agent-legibility problem specifically. Confidence: high.
• Figma internal candidate: Knows the product and the politics but lacks the outside read on context as an architecture problem. Hardest archetype to assess from public signal. Confidence: speculative.

One objection: "Where is the context-system artifact?" The essay is a framework and the Agentic Labs apps are production demonstrations. Neither shows how context passes across five product areas at the interaction-model level, which is what this role is. Counter: the five handoffs are the architectural logic, CarrierIQ demonstrates the pattern in a regulated domain, and TinyFish gives you the production failure modes. Name the gap in conversation as work in progress. It holds up because it's accurate.

PHASE 3 — THE OUTREACH PACKAGE

Contact routing:

• Primary: Loredana Crisan — confirmed expanded product-and-design scope. No direct hiring manager for this role is publicly identifiable.
• Secondary: Noah Levin — confirmed oversight of Figma's Product Design org. Message him if Crisan hasn't responded in seven business days.
• Do not contact first: Matt Fichtner — Design Manager on next-gen design-systems tools, likely reports into this role.

Warm path: No mutual connections surfaced. This is cold, which means the opening question carries the full burden of getting read.

The Persona Decoder read on Crisan's expanded-scope archetype applies: a leader inventing her own remit evaluates candidates against her working model of the organizational failure that created it. Her public signal tells you what that model is — AI making individual work easier and collaboration harder, teams needing shared context and the agency to choose a direction. Mirror it before you introduce anything of your own.

First Contact Message — to Crisan

Subject: Where product intent lives when the system is legible to agents

Loredana — you've described AI as making individual work easier and collaboration harder, and teams as needing shared context plus the agency to choose a direction. That's the problem I've been working on from the other side.

Once components, tokens, and Code Connect make a design system readable to an agent, the system gets more legible. Product intent, usage exceptions, and the judgment about when generated output is ready to leave the canvas don't travel with the tokens. They live somewhere, and deciding where is an architecture problem.

I've written about this as a sequence of five handoffs — moments where authority passes between a human and an agent, each requiring different context to pass along with it. The framework came out of shipping an enterprise agent platform and building three production agentic applications, and it maps onto the Context Platform Pillar.

Could I have 20 minutes on where you see the widest context-passing gaps across Design Systems, Code, MCP, CMS, and Community?

— Juno Chen

Cover Letter

Figma's MCP server makes components, tokens, and code mappings legible to agents. Three decisions underneath that remain human: where product intent lives, how drift between design and code gets governed, and when generated output is ready to leave the canvas. Each needs a designed surface. I've been building the framework for those surfaces as a trust-handoff architecture that maps five moments where authority passes between a human and an agent, each one requiring different context to travel with it.

The closest production proof is Carrier IQ, an InsurTech quote-automation application I built at Agentic Labs and published at junochen.com. Its provenance chain — form completion, coverage selection, approval gate — solves the same structural problem as Figma's canvas-as-review-surface: a user inspects agent-produced context before authorizing a consequential action. Designing that approval gate forced the questions your pillar faces. What context must travel with the agent's output for the human to decide confidently, and what happens when the agent's selection doesn't match an intent the user never stated. CarrierIQ authorizes a policy bind rather than releasing code, but the handoff architecture is the same.

Most recently as Head of Product at TinyFish, I shipped an enterprise web agent platform from 0 to 1 in three months, tracing agent runs in production and working through auditability and governance in enterprise deployments. I moved to product to build AI-natively from zero. I've come back to design because the problem your pillar owns — how intent and judgment travel across design systems, code, and agents — is the design problem I want to work on at this scale.

I'd welcome a conversation about where the hardest context-passing gaps sit across the five areas. Portfolio and published framework at junochen.com.

Resume Emphasis Guidance

Lead with the Trust essay and Agentic Labs; they carry the context-architecture claim this role evaluates against. Second, Alibaba.com (Head of Design and Research, North America; $50B+ GMV, +20% daily transactions, +2.2pt NPS) for enterprise platform scale and structural-gap diagnosis at Figma's level of complexity. Third, Thermo Fisher mySupply (Product Design Director, BCG Digital Ventures; $20M+ margin recovered annually, six pharma partners, 100% adoption) for regulated 0→1 under real stakes. Trim Red Cross and Equinox+ unless asked. Header reads "Most recently, Head of Product at TinyFish (enterprise web agent platform, Series A)" with one line on shipping 0→1 in three months and tracing agent runs in production. Handle the transition in the summary rather than leaving it for the screen: moved to product to build AI-natively from zero, returned to design to apply that depth to a specific architecture problem. Use the posting's vocabulary — context architecture, design systems legibility, agent-human handoffs.

What to Ask in Conversation

"How do you think about the boundary between rules the system should enforce and choices that need to stay open to human judgment, and does that boundary move across the five areas in the pillar?"

Engagement Intelligence

Warning sign: This Director reports into someone who is currently inventing her own role. That's good if Crisan is building the org around a settled mandate. It's destabilizing if the mandate keeps shifting as her scope crystallizes. Ask how she sees the pillar's charter evolving over the next year. The specificity of the answer tells you how much autonomy the role actually has.

What not to say: Don't position as a design-systems specialist. Design systems is one of five areas in the pillar, and leading there signals you're solving a fifth of the problem.

The non-obvious thing: Because the AI-design leadership roles are being filled on overlapping timelines, this Director's influence over team composition and the division of work between pillars is at its maximum right now and declines with each adjacent hire. Raise the sequencing in conversation. Competing candidates will have read the job description and stopped there.

Post-Outreach Signals

Figma is post-IPO with a strong growth trajectory, so three to five business days is a normal response window. A reply that engages the context-passing question is the signal you want — answer inside 24 hours and go deeper. Silence past seven business days: one follow-up on day eight anchored to a new hook, a Figma product update or a Crisan post. If Levin or Fichtner respond instead, the message reached the org and the thread is still live.

PHASE 4 — WINDOW SUMMARY

October 5, 2026.