Observant.

Put user learning
on autopilot.

Observant talks to each of your users one-on-one and keeps learning automatically — following up in the moment, relaying your team's questions live, and pipes user truth straight into the agentic loop.

The problem

Agents do everything now — except learn from users.

Agents write, design, test, and ship. What they can't do on their own is reach a real person and ask how they actually use it, or what they think. Learning from users is still operationally heavy and frictional.

Customer research

Learning from users is the new bottleneck.

Across the ~20 founders and operators we've talked to building with AI, the same gaps came up:

Research is a chore — finding, recruiting, and scheduling the right people.
Random, not systematic — builders rely on signals from social, Discord, or Slack with customers; they're sporadic and unstructured.
They lack the research know-how — to ask the right questions and read the answers.
They want their behavioral data connected, not siloed — the truth sits unread in their analytics.

"I just wish every Tuesday between 1pm and 3pm I could talk to customers — and there's just customers there for me to talk to."Chris, Founder

"All the questions come to me… Some PM says 'I want the result by end of this month' — and we can't cover it. So those questions are in the air. No one answers them."Kai, Researcher

The reframe

Observant completely reimagined how companies learn from users.

The unit of learning shifts — from the study to the individual. Instead of standing up a study when you have a question, Observant learns continuously from each user, and insight rises on its own.

The old way	The new way — on autopilot
inquiry-driven, ad-hoc studies	bottom-up, always-on product discovery
relies on a fixed sampling frame	1:1 at scale
research plan → alignment → recruitment → data collection → insights buried in Notion	runs itself
decoupled from behavioral data; effort to combine sources	all data plugged into your agentic workflow

The product

A direct line between your product team and your users.

Observant automates the decisions of user learning, and it runs on your own users: a bi-directional standing panel — consented, compensated, audited — where the agent holds a continuous 1:1 with each person and keeps a durable memory of them. What grows over time is how much the agent decides: from who, when, and how to ask — to what to ask.

Day one — the short-term goal

The PM brings the question — that's the human part. Everything after is the agent's: who to ask, when, and how — a light exchange or a 10-minute interview, over email or IM. It follows up on its own and returns answers as PRD-ready hypotheses.

The end goal — the long-term vision

The self-evolving product. The question stops needing a human: a funnel drop or a churn pattern opens the conversation, and what comes back flows into the build loop — bugs fixed, tactical design issues surfaced, automatically. Strategic judgment stays with the team.

Landscape

Everyone else hands you a tool to run a study. Observant builds you the Human API.

Category	What it covers	Who operates it	Cadence
Survey tools Typeform · Qualtrics	a narrow execution tool — one methodology	you + a research team + ops	weeks per study · snapshot
AI interview platforms Listen Labs · Outset · Synthetic Users	one phase of execution — ad-hoc; still ops-heavy for study design, internal sampling, recruitment	you + a research team + ops	hours per study · snapshot
Traditional research SaaS Dovetail · UserTesting · Maze · dscout	one phase per tool — still ops-heavy, ad-hoc, and you stitch the pipeline yourself	you + a research team + ops	weeks per study · snapshot
Observant	the whole pipeline — decisions and execution, on your own users	runs itself	continuous

The vision — our big bet

We believe the Human API is a new category.

The belief underneath it all: human signal is the scarce primitive of the agentic era — code got cheap; knowing why your users do what they do didn't.

Many companies will be built in this category. Our entry point is the most acute pain — automated user learning — and from there the panel and methodology compound into the layer every agent queries for human truth.

Where we are now

Customers: We are currently partnering with Parafin (Series C fintech) and Wyze (Series B consumer tech) to build them self-serve user feedback pipelines — with more inbound in the queue.
Build the technical moat — agent-ready feedback — the harness: a decision layer that turns scattered human feedback into structured, testable, agent-ready PRDs.
In the long run — build our own panel — a standing panel of consented humans for high-quality matching, to co-build products with them. Owned human access is the scarce asset in the age of AI.

Team

Xuan Zhao — co-founder & CEO. Xuan has lived the early-startup grind from the inside, and learning from users has been her life's work since — the problem she cares about most. One of the first 60 employees at Robinhood, where she built and led user research through Robinhood's highest-growth years — directly responsible for the 0→1 research effort for most of its flagship products, including Options ($222M, Q4 2024), Cash Management ($296M, Q4 2024), Banking, Web, Passive Investing, and Robinhood Learn. She went on to lead monetization research at pre-IPO Airbnb and Instagram, and stood up research from scratch at SmartNews and Wyze. PhD from the University of Michigan.

Zhifei (Jeffrey) Song — co-founder & CTO. Fifteen years building large-scale backend, ML, and AI systems at startups and public tech companies. At LinkedIn, as a Senior Engineering Manager, he built and led the teams behind recommendation systems, LLMs, and AI platforms serving a billion members. The lesson that brought him here: the hard part isn't model capability — it's building systems that continuously learn from users and improve over time.

Contact: xuanzhao630@gmail.com · jeffrey.listening@gmail.com