Phase 1 — The Moment
Databricks posted this role September 29, one day after Omnigent's managed beta went live (Databricks blog, September 28). The product is in the field without a named design lead, and the posting scopes this hire to the interaction frameworks it does not yet have.
The real mandate: design the layer where developers supervise fleets of coding agents — approving, forking, recovering, intervening from mobile — inside a product whose contextual policies system already converts identical agent actions from automatic to approval-gated as session risk accumulates. The posting calls this person "Omnigent's IC design lead" and names collaboration, approvals, observability, and recovery as the framework surfaces.
Scoring:
| Score | Basis | |
|---|---|---|
| AI centrality | 3 | AI is the core product. Omnigent is an agent orchestration layer. |
| Stage / equity | 1 | Pre-IPO at a reported ~$190B private-market valuation. I could not confirm the round date or how recent the secondary pricing is. Real equity, but the entry price limits upside against Series C–E timing. |
| Design influence ceiling | 2 | The posting assigns framework-level authority. A company-wide VP/Head of Design exists (Ryan Donahue, confirmed via LinkedIn). Both available contacts for this product are PMs, and no public design reporting line for Omnigent surfaced in research. |
| Trajectory | 3 | ~65% YoY revenue growth (most recently reported figure; reporting period not pinpointed). Beta launched four days ago. Three product design postings currently live, all AI-product scoped (careers page, reviewed October 3). |
| Company total | 9 | |
| Total comp | 2 | Posted Zone 1 base tops at $229,150 plus bonus and equity. Cash may approach the floor without reliably clearing it. |
| Scope expandability | 3 | IC design lead for a new product in rapid expansion. No visible ceiling. |
| Craft depth | 3 | Hands-on Staff IC. Interaction frameworks, no reports. |
| AI exposure quality | 3 | Agent supervision, contextual policies, fleet orchestration. This is the design problem. |
| Portfolio value | 3 | Publishable and differentiated: agent fleet supervision at a $190B pre-IPO company. |
| Role total | 14 |
Day 4 of the posting, day 5 of the product. Apply and send the first message today.
Phase 2 — Portfolio Mapping
Intelligence layer check: qualifies. Omnigent's design surface sits between what the agents produce and what the human authorizes. Its contextual policies system is the clearest production implementation of the Trust essay's staged-authorization pattern at any target company so far: a session risk score accumulates as the agent reads sensitive material, and actions that ran automatically early in the session require sign-off once the score crosses a threshold. The agent's behavior hasn't changed, only the accumulated context, and the design problem is making that legible to the person suddenly being asked to approve.
Lead with:
- Trust essay. Watch → Verify → Delegate maps onto Omnigent's supervision model without translation. The argument that users anchor to the worst outcome rather than average accuracy is the operating thesis for a product where agents write production code.
- Brand Pulse. The closest published proof of the observability pattern fleet supervision requires: every signal scored as it lands, urgency and sentiment surfaced without manual review. That's the question this role answers at scale — what does the human watching an agent need to see?
- Carrier IQ. Staged authorization in a regulated domain: a human reviews agent-produced evidence, then authorizes. Same sequence the contextual policies implement. The domains diverge — Carrier IQ authorizes a policy bind, Omnigent authorizes a code change — so state the structural match and stop there. Overclaiming it with engineers in the room costs more than the analogy earns.
TinyFish framing for this audience. Most recently Head of Product at TinyFish, shipping an enterprise web agent platform from 0→1 in three months, working daily with agent traces, session replays, auditability, and governance in customer deployments. Use it as practitioner currency — the reason the Trust essay reads as field-tested rather than theoretical. Not a case study. Does not appear in the portfolio.
Evidence readiness. The mapping on supervision, observability, and staged authorization is the strongest of any role surfaced on the board to date. Two gaps are real.
Developer-product track record. The posting asks for "shipping complex, high-quality developer products." The published portfolio has no developer tools, IDEs, or coding-workflow products. TinyFish is adjacent — an agent platform operated by technical users — but not the same thing. Specific experience with "coding tools, IDEs, agent frameworks, or developer platforms" appears under nice-to-have, not required. The posting treats those as two different levels of qualification, and so should you.
Codebase contribution. The posting asks for a "builder-designer able to make small changes in the codebase." That phrasing is unique to this role among the currently live Databricks design postings, where coding otherwise shows up as prototyping or nice-to-have. Read it as a real capability expectation rather than boilerplate.
The single hardest objection: "Have you shipped developer products?" It is hard because the honest answer is no, and the codebase language signals a team that wants someone who lives inside developer workflow. The closest published evidence is the three Agentic Labs apps: solo-built, full-stack, deployed and maintained by her. Not developer tools, but proof of the working mode the posting describes. Do not try to close this gap before making contact. The buyer can evaluate the mapping faster than you can build another artifact to preempt it.
Phase 3 — The Outreach Package
Contact map. No design reporting line for Omnigent is publicly visible. Ryan Donahue (VP/Head of Design) leads design company-wide, with no public connection to this product or this requisition.
Ridhhima Gupta is a Staff Product Designer "designing for Agents and AI Infrastructure" — scope-adjacent, not confirmed on Omnigent. Her announced Data+AI Summit session, "Designing for Agents: A New Playbook for AI-Native Product Experiences," works from an L0–L5 autonomy framework and decisions about human control and trust surfaces. That vocabulary tells you how the design team frames autonomy internally; expect L0–L5 to be the frame if you reach a design evaluator. Treat her as a tertiary route if neither PM path produces a design connection.
Two substantiated product-context routes follow. Both are product managers, not confirmed hiring managers. They can discuss the problem and route you to whoever owns the design decision.
- Kasey Uhlenhuth — Director of Product Management. Co-authored the Omnigent introduction with Matei Zaharia. Publicly described agent-mediated peer review as a core Omnigent pattern.
- Elise Gonzales — Staff Product Manager, Agent Bricks. Publicly described using Omnigent for her own PM work. Co-authored the Smart Routing article on model-choice overload and underspecified prompts.
Write Uhlenhuth first. She is the more senior product voice and the peer-review post is the better hook.
First Contact Message — Kasey Uhlenhuth
Kasey — your Omnigent post named something I've been circling in my own work: peer review changes shape when the thing being reviewed was produced by an agent. The reviewer needs different evidence, and "approved" means something different when the author can't account for its own reasoning.
I've been designing against that problem directly. I published a framework this year on where trust transfers between a person and an agent, and I built three production agentic systems — brand monitoring, field intelligence, insurance quoting — where the central question was what the supervising human needs to see and when they should step in. Most recently I was Head of Product at TinyFish, shipping an enterprise web agent platform.
I've applied for the Staff Product Designer role on Omnigent. I'd welcome 20 minutes to talk through how the supervision and staged-approval patterns I've shipped line up with what you're building, and who on the design side I should be speaking with.
Resume Framing Note
Open the summary on the Trust framework and the TinyFish build: designer and builder of human-agent supervision systems, published framework for trust handoffs, shipped an enterprise agent platform 0→1. Surface Brand Pulse first, Carrier IQ second. Push Alibaba and the BCG DV work below the fold; this reader did not ask for it. Strip management vocabulary entirely — team building, hiring, org design, function building — because the posting contains none of it and leading with it reads as a candidate applying to a different job. TinyFish in past tense, titled and dated. Do not reach for "developer tools" or "IDE" framing anywhere on the page; the adjacency isn't there and claiming it invites the one question you cannot answer well.
Cover Letter Hook
Omnigent's contextual policies already convert identical agent actions from automatic to approval-gated as session risk accumulates. That is the staged-authorization pattern I designed for in Carrier IQ and formalized in my published trust framework, where the design problem is whether the person being asked to approve understands why the permission changed.
Phase 4 — Window Summary
| Action | Deadline | What degrades without it |
|---|---|---|
| Submit application, send Uhlenhuth message | October 3 | Day 4 of a posting on a five-day-old product. Later contact means being read against people already in conversation. |
| If Uhlenhuth is silent, message Gonzales with the Smart Routing hook | October 7 | Reference the model-choice-overload article and what routing on underspecified prompts does to the interface. Without it, one unanswered message stalls the whole engagement. |
| Follow up with Uhlenhuth or Gonzales; if neither has answered, go to Donahue as company-wide design leader | October 10 |
- Gupta's autonomy framework session: Ridhhima Gupta's announced Data+AI Summit talk on an L0–L5 autonomy playbook for AI-native product experiences will reveal how Databricks' design team internally frames human control over agents — useful vocabulary if you reach a design evaluator.
- Gonzales on routing ambiguity: The Smart Routing article Gonzales co-authored identifies a specific unsolved interaction problem — the system commits to a model-and-harness choice when the opening prompt is least informative — which is a concrete design question to raise if she becomes your second contact.
- Contextual policies as design surface: Omnigent's policy tutorial shows the current UI for session cost budgets, soft-warning approval prompts, and hard-block recovery, but documents no causal audit view or general rollback mechanism — territory this role would likely own.
- Comparator postings for gate calibration: Databricks' other live design roles — Staff Product Designer, AI Products and Sr. Product Designer, AI/BI — treat coding as nice-to-have or prototyping-oriented, which confirms the codebase-contribution language in the Agentic Coding posting is a deliberate product-specific expectation worth testing directly with the team.