Three days a week in New York, or stop here
The Director, Product Design posting leaves no room on geography.
| Location | New York, no secondary location |
| In office | 3 days/week minimum |
| Base band | $320,000–$440,000 + equity |
One wrinkle: Ramp's structured feed tags the role remote-eligible while the same record says Hybrid and the prose names New York only. That's a question for a recruiter, not an opening.
Answer it now. If three days a week in New York is a no, close the card. None of what follows converts. If it's a yes, or a "for the right role," keep reading, because nothing else on the board matches your agentic work this closely.
The trigger
Published June 8, 2026. Still live in Ramp's canonical job feed when I rechecked it on August 1. The role wants someone "building tools and agents, designing memory, shipping PRs." Confidence the posting is open: high. Confidence in what its age tells us: moderate.
The window, and where you are in it
Fifty-four days old. That's past my freshness gate, so carry it with a label attached: live but extended, verify-state.
There are two honest readings. Director searches at this scope run sixty to ninety days routinely, with nothing wrong underneath. Or a first slate came and went, nobody converted, and the committee now knows something in August it didn't know in June. You don't need to resolve which one it is, because both readings point the same direction: the portal has been absorbing inbound for almost two months and is the worst channel available to you. Go at Zaks directly. File the application inside twenty-four hours anyway, so the record exists.
Look at the sequencing. Ramp announced a $750M Series F at $44B on June 4. The posting appeared four days later. Headcount authorized off a round that fresh usually carries a mandate nobody has hardened yet.
What the posting implies about the design org
Ramp has a design function and Zaks has run it for years. The posting describes a shift in what that function is being asked to do: own a product area and a team, stay in contact with customers, get hands into the work to unblock shipping, and help define how design operates once agents perform a share of the labor.
Now read what's missing. The product area is never named. Managing managers shows up as nice-to-have rather than requirement. That pairing, and I'm reasoning from pattern here rather than direct evidence, usually means a company certain it needs senior design judgment aimed at agent surfaces and unsettled about which surface. Scope is negotiable in the first conversation, which almost never happens at Director level. Use it.
Your angle, and why it isn't Alibaba
Lead with the Trust essay and Carrier IQ. One pillar. Alibaba does not open.
That's a Ramp-specific call, not a default for AI companies. Ramp already shipped the inspection layer, and shipped it well. Policy Agent launched review-only, routes the ambiguous cases to a person, permits override, and logs its rationale alongside the policy version it cited. AP Agent memory justifies every populated field by pointing back at the approved bill it learned from. Agent Cards mint a single-use credential bound to one merchant and one amount, dead after a single authorization.
Visibility before the fact, constraint before the fact. What Ramp hasn't publicly resolved is everything after. Split the question into its four verbs and run them live in conversation: can a person inspect what the agent did, interrupt it mid-flight, correct it before commitment, reverse it afterward. Ramp owns the first. The fourth is largely open. Your five-handoff framework gives that seam a vocabulary it currently lacks. Carrier IQ is the artifact that proves you've built into it: insurance quote automation, form completion, coverage selection, a decision gate before bind, in a domain where an unverified number becomes legal exposure, structured to demonstrate all five handoffs.
Alibaba earns one sentence of scale confirmation. $50B+ GMV, +20% daily transactions, +2.2 NPS. It establishes that you've carried trust architecture at volume. It says nothing about designing an agent's undo, and undo is what this role is buying.
Never say "AI trust" in front of Zaks. Say interrupt and reverse, and treat them as two separate design problems.
Who to contact
Diego Zaks, VP Design. linkedin.com/in/diegozaks. His author page confirms he oversees product design. Confidence he is the buyer or sits one step above the buyer: high. The posting names no hiring manager, and I found no evidence of a Head of Design layer between him and the role. Assume direct. If a manager-of-managers surfaces in round two, that person is your real second-round audience.
Public fingerprint. I went looking for a dated statement inside the last twelve months where Zaks says what authority a person keeps when software acts for them. It doesn't exist. His most recent byline is the Midday acquisition post from May 7, 2026, which frames the ambition as design building memorable experiences while AI handles work on the customer's behalf. There's a Fast Company interview from September 2025 and a UX Tools appearance from April 2026.
His most useful published line on human authority is old. Carry the date when you use it:
"Good software design flags situations that benefit from human review while automating the rest."
— Diego Zaks, Ramp blog, November 2021
Quoting something from five years back, dated accurately, is the cheapest available proof that you read rather than skimmed.
Warm path: nothing verified. Cold. Channel: LinkedIn DM, portal within twenty-four hours after.
Outreach messages
Cold message. This is the one you send.
Subject: Where reverse lives once the agent has already paid
You closed the Midday post on AI handling work on the customer's behalf, and back in 2021 you wrote that good software design flags what benefits from human review and automates the rest. Policy Agent defaulting to review-only and Agent Cards scoping to a single use are that idea shipped: inspection before, constraint before. Your support docs are honest about what comes after. An approved card review has no general undo, and card transactions can't be live-approved because authorization happens in real time. I published a five-handoff framework for that seam in June and built Carrier IQ against it: insurance quote automation with a decision gate before bind, in a domain where a wrong number becomes legal exposure. Worth 20 minutes on where interrupt and reverse belong in Ramp's agent surfaces? Both are at junochen.com.
Warm message. Hold it until a path opens inside Ramp design or on the Series F investor side:
[Contact] pointed me to you directly. Ramp posted the Director, Product Design role four days after the $44B round closed, and the way it reads, the open question is authority after commitment rather than inspection, which Ramp has already built. That's the problem I designed Carrier IQ around in insurance and the one I named in a five-handoff framework I published in June. Before that I led the Alibaba.com B2B redesign as Head of Design and Research for North America: $50B+ GMV, +20% daily transactions. Worth 20 minutes?
What not to say. Don't tell Zaks that Ramp needs to build trust into AI. The man shipped review-only defaults, override paths, and scoped agent credentials; that observation reads homepage-deep coming from anyone. Second landmine: don't cast yourself as the person who will build the design function. It exists. Pitch the product area.
Cover letter draft
Ramp has solved the visible half of agent authority. Policy Agent logs its reasoning and the policy version it cited. AP Agent explains why each field was filled. Agent Cards expire after one authorization. Your own support documentation is where the other half stays open: there's no general undo on an approved card review, and card transactions can't be live-approved because authorization happens in real time. The distance between what a person can inspect and what a person can take back is the design problem I want to own.
In June I published Trust Is the New Interface, a five-handoff framework for agentic systems (Intent-Setting, In-Progress, Output Review, Decision Gate, Loop Feedback) sitting on a Watch → Verify → Delegate ladder. Carrier IQ, live at junochen.com, is that framework built: insurance quote automation running form completion and coverage selection through a decision gate before bind, in a regulated domain where an unverified number carries legal exposure. Before that I led the Alibaba.com B2B platform redesign as Head of Design and Research for North America. $50B+ GMV, +20% daily transactions, +2.2 NPS. It started by naming a structural gap between what the platform assumed and what buyers were actually risking.
I'm Head of Product at TinyFish now, an enterprise web agent platform (Series A). I moved into product to build AI-natively from zero and shipped three products in the first three months. Most of my days sit inside agent traces, attribution, and the auditability bar enterprise buyers set before they'll let anything run unattended. I'm coming back to design because I want that depth aimed at one high-stakes vertical instead of a horizontal platform, and because craft is the part of the job I miss.
The posting says design at Ramp means building tools and agents, designing memory, shipping PRs. I'd like 30 minutes on which product area you'd want this to start with, and what reversal looks like once agents are the ones transacting. Work is at junochen.com.
Resume emphasis
Craft depth scores 3: the posting asks for code prototyping and shipped PRs outright. AI exposure quality scores 3. Those two numbers fix the order.
- Trust essay first. Title, date, one line on the five handoffs. It's the credential that says you named the problem before they wrote the posting.
- Carrier IQ second, described by mechanism, never as "AI project." Regulated quote automation, coverage selection, decision gate before bind.
- Alibaba third, compressed to three numbers.
- Thermo Fisher fourth, kept for one reason only: the single human gate at batch QA release. That's a reversal-boundary decision inside a regulated system, and it's the conversation you want to be having.
- Red Cross and Equinox+ down to a line each. Neither pays rent at Ramp.
- TinyFish in the header. "Head of Product, TinyFish (enterprise web agent platform, Series A)," plus one line covering three products in three months and daily work in agent traces. Context. Not a case study.
Who else is in this pipeline
AI-native design leader (confidence: high). Frontier lab or a well-funded agent startup, carrying better brand recognition than yours. The objection they create: your AI proof is self-built. Counter: say Carrier IQ, Thermo Fisher, and Alibaba in a single breath. Regulated 0→1 with $20M+ in annual margin recovered, plus a $50B+ platform. Their advantage is the logo; yours is having designed where the outcome carried real cost.
Enterprise design director (confidence: high). Larger org, more headcount managed. The objection: you haven't run managers of managers. Counter: the posting files that under nice-to-have. Don't volunteer the gap. Talk about owning the area.
Staff or Senior Staff IC from a late-stage fintech (confidence: moderate). Deep craft, fluent in payments primitives. The objection: you don't know the domain. Counter: Carrier IQ is insurance, and the authority question there is structurally the same question.
Where you're genuinely exposed. Agentic Labs is solo-built with no adoption, revenue, or customer-impact numbers attached, and against your enterprise cases that asymmetry is real. A reviewer will feel it whether or not he says it. Name it first, then draw the distinction that actually holds: these run on real APIs with live agents against real data, not prototypes, not mockups. Then go straight to Thermo Fisher for the metrics.
Post-outreach signals
Growth-stage cadence applies. A reply inside 3–5 business days is normal. Silence past 7 means a deep slate or a search that's moved past you. Follow up once on day 8, hooked to something new, a release or a post. Never a check-in. If he engages the reversal question rather than routing you to a recruiter, answer inside 24 hours and go deeper on Thermo Fisher's single human gate. A recruiter screen booked off first contact means the search is moving fast; drop everything and prep.
The non-obvious thing
Almost nobody in this pipeline will have read Ramp's transaction review support doc. It says a single transaction can be moved back one approval step only where a "Revert to previous approver" control exists, and that control isn't universally available. It also says card transactions cannot be live-approved, because authorization happens in real time, and it steers administrators toward category restrictions and spend limits instead.
Read that as Ramp documenting its own tradeoff in public: prevention chosen over interruption. Every candidate walks in with an opinion about AI and trust. You can walk in with the sentence where the company already drew the line, and ask whether the line survives agents doing the transacting.
Warning signs
Equity math at $44B. The Series F was announced as $750M, all primary, with no employee tender disclosed. The prior round, November 2025 at $32B, did include one. A grant struck at $44B needs a large multiple before it's life-changing, and no public source discloses liquidation preference or strike price. Price the cash accordingly.
Scope is stated, not guaranteed. "Room to take on broader leadership over time." No named area, no team size, no expansion milestone.
Those two interact. You'd be relocating for a base number and an unpriced option. At $320K–$440K that can still be the right trade. Just make it on purpose.
Ask these before you accept:
- What's the liquidity precedent for employees, and was the November 2025 tender a one-off?
- Which product area does this role start with, and how big is that team today?
- What's the written expansion milestone, and what triggers it?
None of it is disqualifying. Send the message.
- Agent identity architecture: Ramp models every agent as an actor attached to a human sponsor so downstream systems retain both identities, which is the closest thing Ramp has published to an accountability design principle and worth reading before any second-round conversation — Ramp's builders post on agentic identity is the primary source.
- The Zaks interview to watch: His April 2026 UX Tools episode on design at the most AI-installed company covers Ramp's internal AI adoption and disappearing interfaces, and its chapter markers will tell you which of his current framings to mirror in a follow-up message.
- Brex as the parallel bet: The Staff Product Designer, AI role at Brex maps almost exactly to your Watch → Verify → Delegate ladder and CDO Matt Bango publishes his own email for candidates, but Capital One closed its acquisition on April 7, so treat the equity thesis as gone and the role quality as intact.
- Vanta's quality-gate precedent: If you want a working model of design owning AI release authority rather than AI surfaces, Vanta's published account of how it builds AI products describes task-specific evaluations and expert-calibrated reviewers as shipping gates — useful vocabulary for asking Zaks what design can actually stop.