Leading AI-first product design through three roles and two company re-orgs.
Six years at Zendesk, most recently building and leading the design team behind Zendesk Copilot — the human side of an AI portfolio that passed $200M ARR in 2025.
On this page
01 · The setting Senior Manager, Product Design — Copilot · Aug 2024 — Present
Zendesk bet the company on AI. My team designs the human half of that bet.
Zendesk's AI comes in two product lines: autonomous AI Agents that resolve customer issues end-to-end, and Copilot — AI that makes the people running service better at their jobs. My team owns Copilot's design: the flagship experience for support agents, a company-wide assistant for admins, and the adoption work that turns purchases into usage.
of the Copilot SKU
attributable to Copilot
projected to double in 2026
The payoff moment: at Relate 2026, Zendesk's annual conference, most of my team's roadmap went GA on one stage.
02 · Admin Copilot GA May 2026 · included in core plans
My team shipped the horizontal capability that defines how Zendesk is run in the AI era.
Admin Copilot is one conversational assistant shared by every Zendesk administrator, across the entire product. It watches how an account performs, flags what's broken, explains it in plain language, and fixes it with approval. Not a feature — a platform layer, and my team designed it.
- We designed the framework, not just the screens — product teams across the company plug their capabilities into patterns my team owns. At GA it launched with 70+ types of proactive recommendations from all over Zendesk, and still feels like one product.
- We made AI trustworthy enough to touch production — the preview-and-approve contract: the assistant can change live configuration because every change is shown and confirmed first. This pattern is what let a conversational AI ship to admins at all.
- The rollout was a distribution decision — unveiled Oct 2025, opened to all customers Mar 2026, GA at Relate in May 2026 bundled into core plans rather than sold as an add-on. That made it infrastructure, not an upsell.
03 · Auto Assist GA Oct 2024 · flagship of the Copilot SKU
AI that makes support agents measurably faster — with a human still in charge.
Auto Assist sits inside every support ticket: it reads the conversation, proposes next steps, drafts replies, and executes actions with the agent's approval. The UX bet: keep the human in charge, but collapse "what do I do next?" to nearly zero. The hard design problems were trust and control, not visuals:
- Configuration in prose, not builders — the AI follows "Procedures": playbooks written in plain language ("check the order status, then offer a refund up to $50") instead of decision trees. This concept was born on my Knowledge team before Copilot existed.
- Control is a dial, not a switch — admins choose which actions the AI may take without asking, per situation. Autonomy is granted gradually, as trust is earned.
- Hybrid human–AI workflows — the AI handles what it can and hands the agent a checklist for what it can't, inside the same flow.
- Interrupt only when confident — suggestions ship only above a confidence bar, and the system learns in real time from how agents respond. Fewer, better interruptions.
04 · Adoption & PLG 2025 — 2026
Buying AI is easy. Using it isn't. We made adoption a design problem.
Copilot was selling, but activation lagged: value stayed invisible until an admin invested hours in setup. We redesigned the entire path from "purchased" to "habit":
- Setup went from authoring to reviewing — instead of asking admins to write playbooks from scratch, the AI now drafts its own configuration from the customer's existing internal content. The single biggest cut to time-to-value.
- First run ends with something working — a redesigned onboarding flow that produces a live, working assistant in the first session, not a settings tour.
- Value before configuration — a mode that gives useful suggestions out of the box, no tuning required, so the product proves itself before asking for effort.
- Growth loops woven into the product — a weekly digest that re-engages admins, and recommendations that double as a discovery surface for capabilities customers own but haven't turned on.
05 · The team up to 9 direct reports · junior to principal
A deliberately senior team — and people who outgrew it.
- Promoted one of my people into management — they now lead the Knowledge design team, whose work headlined Relate 2026 alongside ours. The best performance review a manager gets.
- Ran a senior-heavy bench — two principal designers plus lead, senior, and junior. The mix you need when the hard problems are platform contracts and patterns, not screens — and a management craft of its own: principals need scope and air cover, juniors need problems that stretch without breaking.
- Kept the team whole through two re-orgs — retained and shipping through every structural change.
- Supported the wider org — standards, patterns, and rituals used across a 40+ designer organization.
How the team runs
Shipping through re-orgs and a platform shift takes more than a roadmap. These rituals keep a distributed team sharp and synced with the wider org.
Design crit weekly
Work-in-progress only — polished work is too late to change. Crit the work, not the person; leave with a decision. Our core AI patterns were forged here before they had names.
Team kickoffs per initiative
Every big bet starts with the whole team in one room: the problem, the customer evidence, what "great" looks like — and what we're explicitly not doing.
Onsites 1–2× a year
Concentrated time for the work that resists remote: messy whiteboards, roadmap arguments, and the trust that makes async work the rest of the year.
Show & tell regularly
Shipped work gets shown — to the team, partner teams, and the wider design org. Part celebration, part accountability.
07 · Four years of shipping 2022 — 2026 · from public announcements
I've led this team since before the GenAI wave — long enough to watch a few machine-learning predictions grow into an AI product line:
Machine-learning triage launches
Account-specific models predicting the intent, language, and sentiment of every incoming ticket — AI before the GenAI wave, and the substrate everything since is built on.
Zendesk's first generative AI features
Summarization, tone shift, and reply drafting for agents. Our first UX vocabulary for drafts, uncertainty, and edit-in-place.
Copilot unveiled at Relate
The human-augmenting track of Zendesk AI gets a name and a stage, alongside autonomous AI Agents.
Auto Assist goes GA
Proactive suggestions, approved actions, and plain-language playbooks. Early customer result: agents tripling throughput, from 40 to 120 tickets per shift.
From launch to product
Auto Assist gains actions in external systems (Jira, Slack), agent feedback loops, pre-approved actions, and hybrid workflows that hand tasks between AI and human — the release cadence of a team compounding on a launch.
Admin Copilot unveiled
The copilot pattern goes horizontal: one assistant for every administrator, diagnosing issues in plain language and fixing them with approval. Zendesk's AI portfolio exits 2025 at ~$200M ARR.
Relate 2026 — the big GA
Admin Copilot goes GA, bundled into core plans. Agent Copilot goes GA, now generating its own configuration from a customer's internal sources. Knowledge Copilot — descendant of my former Knowledge team's work — enters early access.
Adoption, redesigned
New onboarding, AI-generated setup, and confidence-based suggestions that learn from agent behavior. The PLG chapter ships.
08 · Earlier roles 2020 — 2024
Senior Manager, Product Design — AI & Knowledge · Oct 2022 — Aug 2024
I took over Zendesk's AI design team before the GenAI wave, when "AI" meant classifiers, not conversations. This role built the groundwork Copilot now stands on:
- Scaled ML triage from experiment to infrastructure — custom intents, more industries and channels, predictions made legible enough for admins to trust with routing.
- Shipped Zendesk's first generative AI features (2023) — and invented the UX patterns for uncertainty, citation, and revision that Copilot later scaled.
- Owned Knowledge & retrieval — the content every AI answer is grounded in. The "Procedures" concept — operational knowledge written once, in plain language, executed by AI — was born on this team and became Auto Assist's backbone.
- Owned the invisible surfaces — intent taxonomies, training-data curation, model-performance dashboards.
Product Design Manager — Zendesk Sell · May 2020 — Oct 2022
Led the design team for Zendesk Sell, the sales CRM Zendesk acquired (Base CRM) — a standalone product inside a company whose flagship was Support. Equal parts product craft and organisational navigation.
- Owned the design roadmap for a full B2B SaaS CRM
- Integrated Sell with the Zendesk suite, partnering with engineering and product
- First step into people management — rituals, standards, and performance for a small team