Superhuman · AI product · Mid · 45 to 60 minutes

How should Superhuman surface internal research?

Superhuman is an email app for individuals and teams that emphasizes speed, workflow, and AI assistance. Answer this product management case and get it graded line by line against a weighted rubric.

Problem framing
AI product strategy
Prioritization
User insight
Metrics

The situation

You are a PM on Superhuman Go. The Head of Product and a research lead ask you to improve how internal UX and marketing research influences product and go to market decisions. Superhuman already has a growing set of interview summaries, message tests, and other reports, but teams rarely reference them when writing strategy docs, product briefs, or messaging. Leaders worry that important claims are being made from memory or intuition, yet researchers say the work exists and is hard won. Because Superhuman is investing in AI capabilities, some leaders want an AI powered approach rather than another internal docs project. Others are skeptical, citing trust, privacy, and the risk of irrelevant interruptions in the writing flow. You need to recommend what problem to solve first, whether AI is the right lever, and what a first release should look like. The proposal should be narrow enough to ship in one quarter and clear enough that engineering, design, and research can align on it.

What to produce

A 1 to 2 page product strategy memo.

Your answer must

State the primary problem to solve, the target user, and one or two decision moments to focus on first.
Diagnose likely reasons research is underused, and call out key assumptions and evidence gaps.
Compare at least three approaches, including at least one non-AI option and one AI assisted option, with clear trade-offs.
Recommend an MVP with a simple user flow, how answers would be grounded in source research, and how permissions would be handled.
Define success metrics, a validation plan for the first quarter, and the biggest product risks.

Constraints

Use existing internal research documents and tools as the starting point, do not require a full repository rebuild before launch.
Do not depend on adding research headcount or a manual concierge process as the core solution.
The first release must be feasible for one product area within one quarter.
Any AI output must respect existing document access permissions and cite underlying sources.

Format

Use sections for problem, users, options, recommendation, MVP, metrics, and risks.

How Superhuman actually played it

A Superhuman team explains why a better research repository is not enough: people must recognise they need research, leave their workflow and know what to ask. They built an assistant called IRIS that surfaces relevant UX and marketing research in the flow of work, using a connector for access and an agent for action. A model for turning a static knowledge base into a proactive AI feature.

Reduce the effort of finding evidence by bringing it to where decisions are written.
Separate access (connector) from behavior (agent) when designing AI features.
The failure of repositories is user behavior, not storage, so design around the behavior.
Start from a clear moment of need, such as writing a strategy claim.

Built from Research at Your Cursor: Building a Proactive Insights Agent on Superhuman Go by Josh Williams, Superhuman.