Trigger
Headway posted a Staff Product Designer, Provider Practice Growth role (Ashby, live as of September 2026). The build mandate: an AI front desk that handles patient inquiries, booking, and follow-up on behalf of healthcare providers. The design problem underneath it is what happens when the system acts with confidence on a wrong answer — a real patient gets misrouted, and the provider's practice reputation takes the hit before anyone reviews the output. Confidence: high. The posting is live, and Poses's hiring language about agents that "know when to hand off to a human" corroborates the build direction.
Window
~3 weeks remaining. Staff IC roles at growth-stage companies this size run 6–8 week searches. You are mid-window. Ellen Dong's departure as Head of Design (left for Hera, late August) could slow evaluation if the reporting layer is still being resolved, or accelerate it if Tiangco has full hiring authority. Move this week regardless.
Don't hold outreach for forward-looking artifacts to publish. Move now with what's live at junochen.com.
What this trigger probably means for the design org
Headway's design leadership layer is thinner than it was a month ago. Dong left. No replacement has surfaced on Headway's team page, blog, or design community directories. Roland Tiangco, Director of Provider Product Design, is the closest design leader to this role — Poses introduced him as responsible for onboarding, growth, and clinical workflows, which maps almost exactly to "Provider Practice Growth." Whether Tiangco owns this hire outright or whether it routes through an unreplaced function layer is a question you answer in conversation, not from public signal.
The front desk itself is a build mandate. Headway's public AI precedent is Scribe — clinical documentation where the clinician reviews and signs every AI-drafted note before it goes anywhere. The front desk inverts that model. The provider is not in the loop at the moment the agent acts. A patient calls, the system responds, the provider sees the result after. That inversion is why this role exists.
Your angle
Lead with Thermo Fisher mySupply. The front desk is a regulated multi-party workflow where the system acts on behalf of one party (provider), affects another (patient), and the consequences land before review. Thermo Fisher — six pharma partners, nine sites, $20M+ margin recovered, 100% adoption, completed as Product Design Director at BCG DV — is the direct structural match. You designed decision gates for a supply chain where one wrong routing decision cascades across partners who never see each other's data. The connection to Headway's front desk: both systems require designing for the failure scenario where the automation proceeds with confidence on bad information.
Secondary: Allē. Headway's provider-patient split mirrors Allē's provider-consumer dual-surface architecture (30M members, 3.2× redemption, also BCG DV). Use it to show you've designed for the party who delegates and the party who lives with the delegation outcome.
The Trust essay's Decision Gate handoff maps directly to Headway's own stated principle. They already say "Scribe drafts; it does not decide." The front desk needs that same principle applied to a surface where the provider isn't present to decide.
Competitive landscape
Three archetypes are likely in this pipeline (confidence: moderate — inferred from role level, company stage, and domain):
- Healthcare-domain designers (Oscar, Ro, Hims pedigree): They have the vertical. They likely lack AI-native production depth and published trust-design thinking. Your vulnerability against them is direct: no healthcare experience on your resume. The counter: the structural match — regulated multi-party workflow where failure cascades across parties who don't see each other — is more specific than generic healthcare familiarity. The outreach already leads with this.
- AI-native ICs (startup-tier): Model fluency, but unlikely to have regulated multi-party proof or a published framework for agentic trust in high-stakes environments. Thermo Fisher and the Trust essay are your differentiators.
- Growth-stage senior ICs (Ramp/Gusto tier): Right company stage, wrong domain. No healthcare or regulated-environment proof.
The healthcare gap is the objection most likely to surface. When it does, lead with Thermo Fisher's six-pharma-partner constraint architecture. The structural problem transfers more precisely than domain familiarity alone.
Who to contact
Primary: Roland Tiangco — Director of Provider Product Design, Headway. His remit (onboarding, growth, clinical workflows) is the closest public match to this role's scope. Identified in Poses's February hiring post.
- Public fingerprint: Thin. No recent LinkedIn posts, conference talks, or published writing in the last six months. His public signal is limited to Poses's introduction. Because there's no public language to hook on, the outreach message opens on the design problem rather than his words. If he posts anything between now and send, swap the hook to reference it.
- Warm path: None identified. Cold outreach.
- Channel: LinkedIn DM. Reference the provider growth mandate specifically — he'll know immediately whether this role is his.
Secondary: Jake Poses — VP Product (product management, product design, product marketing). The executive sponsor. Use only if Tiangco doesn't respond within 5 business days, or if you learn Tiangco doesn't own this search.
- Public fingerprint: Active on LinkedIn. Posts about agents that "know when to hand off to a human", hiring for "expert-labeled ground truth instead of vibes," and wanting providers to "know, trust, and choose Headway."
- Warm path: None identified. Cold outreach.
- Channel: LinkedIn DM. Hook on his own language about handoff boundaries.
Outreach messages
Cold message to Tiangco:
Roland — the Provider Practice Growth Staff Designer role caught my attention because of the problem underneath it: what happens when the front desk is confident and wrong, and the provider isn't in the loop until after the patient has already been affected.
I designed that constraint architecture at Thermo Fisher (as Product Design Director at BCG DV) — decision gates across six pharma partners where one wrong routing cascades before anyone reviews the output. $20M+ margin recovered, 100% adoption, because the system surfaced its boundaries before it acted.
Would 20 minutes on how that translates to provider-facing automation boundaries be useful? junochen.com has the full case and a framework essay on agentic trust design that maps to this problem directly.
Cold message to Poses (fallback):
Jake — your framing about agents that "know when to hand off to a human" is the exact design problem I've spent the last two years on, most recently shipping an agentic platform from 0 to 1 as Head of Product at TinyFish.
The Staff Provider Practice Growth role maps to work I led at Thermo Fisher (BCG DV) — regulated multi-party workflows where the system acts on behalf of one party and the consequences land on another before review. Six pharma partners, $20M+ margin, 100% adoption. The design problem was the same: what does the interface owe the user when the automation proceeds on bad information?
I'd welcome 20 minutes on how that constraint thinking applies to the front desk. The case and a trust-design framework are at junochen.com.
What not to say:
- Do not reference Ellen Dong's departure or the Head of Design vacancy. It signals you're reading org instability, not the role.
- Do not frame the AI front desk as if it's shipped. It's a build mandate. Talking about "improving" a product that doesn't exist yet tells them you didn't read carefully.
Cover letter draft
Headway's front desk introduces a design problem that Scribe doesn't have: the provider isn't in the loop when the agent acts. A patient gets misrouted or a follow-up falls through, and the provider's practice absorbs the reputational consequence before anyone reviews the output. The question is how the interface makes the system's limits visible to both parties in time.
I designed that constraint architecture as Product Design Director at BCG Digital Ventures on Thermo Fisher mySupply — a regulated multi-party supply chain where six pharma partners across nine sites depended on decision gates that surfaced automation boundaries before the system acted. One wrong routing cascaded across partners who never saw each other's data. The result: $20M+ in margin recovered annually, 100% partner adoption in 12 months. The front desk faces the same structural problem — the provider delegates, the agent acts, the patient experiences the result — and the design has to protect all three parties from a failure that looks like it came from the provider, not the platform.
Most recently as Head of Product at TinyFish, I shipped an enterprise agentic platform from 0 to 1 within three months, working daily with agent traces, auditability, and governance in production. I returned to design because I want to apply that technical depth to a specific, high-stakes vertical — not continue building a horizontal platform.
The full Thermo Fisher case, three live agentic AI demos, and a published framework on trust design for agentic systems are at junochen.com. I'd welcome the chance to discuss how that work maps to the front desk's automation boundaries.
Resume emphasis guidance
- Lead with Thermo Fisher mySupply — regulated multi-party workflow, decision gates, 6 pharma partners, $20M+ margin, 100% adoption. Frame as: decision gates for regulated multi-party automation where failure cascades before review.
- Follow with Allē — dual-surface (provider + consumer), 30M members, 3.2× redemption. The provider-patient split maps directly to Headway's two-sided model.
- Agentic Labs third — Carrier IQ (regulated industry + agentic automation) demonstrates the five handoffs in production. Trust essay provides the conceptual framework.
- Forward-looking artifacts: If Human-Agent System Design artifacts are published at junochen.com, add them after Agentic Labs — they show you're extending into frontier healthcare AI design thinking. If they're in progress, you have a timing call: the 3-week window is tight enough that waiting to publish risks missing it. Move now with what's live and bring the artifact into conversation if it ships before the interview.
- TinyFish as bridge — "Most recently, Head of Product at TinyFish (enterprise web agent platform, Series A). Shipped agentic platform 0→1 in 3 months. Agent traces, auditability, governance in production." One line in the header section.
- Alibaba below the fold — enterprise scale matters here but it's not the lead story. Move below Thermo Fisher and Allē.
- De-emphasize Red Cross — mission-driven but not healthcare (disaster relief, not clinical). Keep it on the resume but don't feature it.
The non-obvious thing
Headway's engineering team published a detailed account of how they evaluate AI-generated clinical notes. The key line: the boundary between an accurate summary and an unsupported clinical assumption can be "a single clause." They run sentence-level checks, statistical gates, precision/recall on the grader, expert-written risk rubrics, and human audits of notes that passed automated review — because reviewing only known failures can't reveal what the automated judge missed.
This evaluation architecture already exists for Scribe. The front desk will need its own version. If you can reference this evaluation philosophy in conversation — the principle that passing automated review is insufficient — you'll signal preparation depth that competing candidates almost certainly won't match.
Post-outreach signals
Headway is growth-stage (Ramp/Gusto/Headway tier):
- Response within 3–5 business days from Tiangco = normal. If he engages the failure-scenario framing specifically, the hook worked. Respond within 24 hours.
- Silence past 7 business days = either in process with other candidates or the role routes differently than public signal suggests. One follow-up at day 8 referencing a new signal (a Headway product update, a Poses post). Then redirect energy.
- Recruiter screen scheduled on first contact = fast-moving search. Prioritize prep immediately — walk the Thermo Fisher case with the front desk analogy rehearsed.
- If Tiangco redirects you to Poses or someone else = useful intelligence about the actual reporting chain. Treat the redirect as signal, not rejection.
- Never follow up more than twice before closing the loop internally.
Warning signs
Comp ceiling. The $212K–$265K range crosses the $250K cash threshold only at the upper end. At Staff IC level, equity matters — but the last public valuation ($2.3B, Series D, July 2024) is over two years old with no subsequent round reported. The equity could be meaningful if Headway is tracking toward an IPO at a higher valuation, or it could be paper. Ask about the equity refresh cycle and the last 409A valuation early — before you invest in a multi-round process where the comp doesn't land.
The missing Head of Design. Dong left. No replacement is visible. That could mean Tiangco and other directors absorbed the function and have real authority — a growth signal for a Staff designer who wants influence. Or it could mean the design function is in a holding pattern with no executive advocate. The prior battlecard flagged the three-party delegation problem; the Issue #10 update first noted Dong's departure. The structural question hasn't resolved. Ask Tiangco directly: who does design report to now, and is there a Head of Design search open?
- Headway's Scribe evaluation architecture: Their engineering team describes how sentence-level checks, statistical gates, and human audits of passed traces work together to catch unsupported clinical assumptions — study this before any interview because the front desk will need an equivalent system.
- Poses's provider lifecycle language: His September hiring post describes wanting providers to "know, trust, and choose Headway" and frames patient lifecycle as the journey from first touch to sustained care — vocabulary worth mirroring if the conversation moves beyond automation boundaries.
- Headway's Tezi acquisition framing: The announcement welcoming the Tezi team describes "human-centered AI" as pairing human judgment with AI agents inside complex workflows, which is the company's clearest public statement of design philosophy for agentic products.
- Jenna Leonardo's AI-sprint model: A Headway Director of Product Design has been publicly discussing AI-accelerated design sprints used with the patient team — worth tracking because her methodology signals how Headway's design org is already integrating AI into its own practice, not just its products.