Opportunity Readiness Brief: Amplitude, Head of Product Design

PHASE 1 — THE MOMENT

Gab Menachem joined Amplitude as CPO on April 14. He owns product, design, and growth. Ninety-six days in, a Head of Product Design posting appeared on Greenhouse (updated July 7, 12 days ago): player-coach, 15-person team, spanning Core Amplitude, Statsig, and Wave. (See Timeline Reconstruction for the full event sequence and window prediction.)

The real mandate: Wave's deep-dive names the human-in-the-loop funnel as "the hardest part." 11% end-to-end ship rate against a ~50% "would do" rate. The system finds things worth acting on. The designed surface between "Wave found this" and "the team ships it" is losing 80% of the value. Three product surfaces. Autonomous design pods. A trust handoff problem running at platform scale with no design leader shaping it.

The org signal you cannot resolve from outside. Kim Lenox is publicly visible as VP of Design at Amplitude (company blog, December 2024; secondary directory traces in 2026). The posting names no reporting line. Under Lenox, you are pitching a senior execution seat with a VP buffer above. Replacing Lenox, you are pitching the top design role reporting directly to Menachem. I cannot confirm which from public signal. Ask about reporting structure in the first ten minutes. Your entire positioning shifts on the answer.

PHASE 2 — PORTFOLIO MAPPING

Intelligence layer check: confirmed. Amplitude's core product surfaces intelligence for human decision-making. The Trust essay's five handoffs map to Wave's published architecture with unusual directness. Each handoff:

• Intent-Setting → Users give Wave "a few sentences about a product area" and Wave maps journeys and objectives. You solved this cold start problem at TinyFish. Blank input box. First-time users bouncing before they reached value. You redesigned the intent-setting layer so users could reach a working agent without prior context. That is §01 of the Trust essay, in production.

• In-Progress → Scoped agent swarms inspect underperforming metrics, feedback, session replays, and traces. Each agent returns candidate problem statements cited and tagged with source. The design question: what does a team see while Wave is thinking, and does visibility during analysis build or erode confidence in the output? You work this daily at TinyFish, tracing agent runs in LangSmith in real time, diagnosing where signals break down between what an agent did and what a user experienced. Making the in-progress state legible is how you catch failures before they reach the output layer.

• Output Review → Wave produces problem statements, specs, and experiments with source backlinks (chart IDs, replay IDs, feedback IDs). Wave's own team acknowledges citations can go stale and session replay analysis is noisy. Evidence and provenance beside every surfaced opportunity, with credibility problems still unsolved.

• Decision Gate → A quality judge and ML classifier predict whether a human would approve or reject. Humans set the autonomy threshold. The 50% "would do" vs. 11% shipped gap lives here. Same economics as Thermo Fisher's exception gate: at Thermo Fisher, exceptions caught a week earlier cost 3–5× less to resolve than those surfaced at the delivery gate. Catching a bad recommendation at the decision point is dramatically cheaper than catching it post-ship. Wave has the same cost structure in a different domain.

• Loop Feedback → Wave captures approvals, rejections, plan overrides, freeform notes, and thumbs feedback, distills them into memories for the next run. Carrier IQ's run-1-misclassification-shapes-run-5 pattern, running at platform scale.

These five handoffs don't analogize to Wave's architecture. They describe it.

TinyFish framing: You use product analytics, session replays, and LangSmith agent traces daily to make design decisions at TinyFish. You are Amplitude's user. You live the intelligence layer design problem from the practitioner side, by hand, in production. When you describe Wave's commitment point, you are describing a problem you already work inside.

The objection. The posting says "we'll ask what you built and what you learned" from AI-native experiences. Menachem founded an AI company (Loom Systems, acquired by ServiceNow 2020) and scaled ServiceNow's AI operations to $1B+ in annual revenue. He will probe whether your AI work is production-deep or portfolio-decorative. He has seen enough of the latter to spot it fast. The answer: Three shipped Agentic Labs demos with real APIs and live agents across enterprise verticals (InsurTech, CPG, brand intelligence), all published at junochen.com. A Trust essay documenting the design problems those demos surfaced. And TinyFish: Head of Product at an enterprise web agent platform, 3 products shipped in 3 months, agent traces and governance challenges in production daily. The depth is operational, not decorative.

PHASE 3 — THE OUTREACH PACKAGE

WARM PATH ASSESSMENT

Menachem's public LinkedIn activity is recent (Agent Analytics, MCP, and Wave posts visible in profile preview), but full post content was not accessible in this research pass. Before sending, check three things: (1) mutual LinkedIn connections between you and Menachem. If 3+, reference them in the close, not the opening. (2) Any Menachem post within the last 4 days. If yes, hook on that instead of the investor release language below. Recency makes cold feel current. (3) Shared connections through ServiceNow, Loom Systems, or Amplitude's investor network. If no warm path exists, send cold. The investor release language is strong enough to carry it.

FIRST CONTACT MESSAGE

Subject: Wave's commitment point problem

Gab — your description of Amplitude as the "intelligence layer" for what happens after release landed with me because I've been working the same problem from the practitioner side. At TinyFish I use product analytics, session replays, and agent traces daily to decide what to build next. The hardest design moment is the five seconds after a team sees what the system found and has to decide whether to act on it, redirect it, or ignore it.

I published a framework for exactly this: five handoffs between what an AI surfaces and what a human commits to. I've shipped it across enterprise platforms at Alibaba ($50B+ GMV, +20% transactions) and regulated supply chains at Thermo Fisher ($20M+ margin recovered). The decision gate design problem Wave is solving at 11% ship rate vs. 50% intent is the same architecture I've been building against for three years.

Would 20 minutes make sense to walk through how the five handoffs map to Wave's current surfaces? I'm available Thursday or Friday this week.

— Juno Chen

RESUME FRAMING NOTE

Lead the summary with "intelligence layer design." Not "AI products." Surface the Trust essay and its five handoffs framework first. Then Alibaba's transaction and NPS outcomes for enterprise platform credibility at scale. Then Thermo Fisher's exception-first architecture for the decision gate in a regulated context. Position TinyFish as: "Currently Head of Product at TinyFish, an enterprise web agent platform (Series A), building and deploying AI agents in production." Subordinate Equinox+ and Allē to a single supporting line. Avoid "design systems" and "design ops" language entirely. The posting explicitly says this role is neither. Mirror the posting's own framing: "autonomous pods" and "coherence across surfaces."

COVER LETTER HOOK

Wave's deep-dive identifies the human-in-the-loop funnel as the hardest unsolved problem, with an 11% ship rate against 50% intent. I've spent three years designing the five handoffs where that gap opens and closes: between what an AI system surfaces and what a human commits to, shipped across enterprise platforms, regulated supply chains, and the agentic products I build daily at TinyFish.

PHASE 4 — WINDOW SUMMARY

July 22, 2026.