Grammarly · AI product · Mid · 60 to 90 minutes

Grammarly whole document rewrite quality bar

Superhuman makes Grammarly, an AI writing assistant and document workspace that helps people draft, rewrite, and improve communication across apps. Answer this product management case and get it graded line by line against a weighted rubric.

User segmentation
Product strategy
Metric design
Prioritisation
Experiment design

The situation

Superhuman's Grammarly product team is deciding whether to expand whole document rewriting from a user invoked tool into a more autonomous workflow inside Grammarly's document surface and Assistant. Today, users can already ask Grammarly Assistant to rewrite full text on request, and Grammarly also offers agents such as Proofreader, AI Grader, and Reader Reactions. Over the last two quarters, support tickets, user interviews, and session logs show a pattern: people want faster ready to send outcomes for long emails, cover letters, essays, and reports, but they also complain when larger rewrites flatten their voice, change facts, or remove needed nuance. The VP of Product has asked you for a proposal because the company is planning the next half of AI writing investments, and leadership wants a clear call on where autonomy should stop. Assume the following numbers are illustrative: 18 percent of active AI sessions include a rewrite request, 31 percent of users who try a full rewrite do not accept the output, and meaning change is cited in 42 percent of rewrite complaints. Your job is to define the product direction and the quality bar, not just ship a bigger button.

What to produce

A 1 to 2 page product proposal.

Your answer must

Identify at least two target user segments or use cases, and name one segment or use case you would deprioritize first.
Recommend one product approach for expanding whole document rewrite, and explain why that level of autonomy is the right choice now.
Define a quality bar with at least four measurable dimensions, including meaning preservation and voice retention, and say how each dimension would be measured.
Provide an MVP scope for the next two quarters plus one later phase, and list at least two explicit non goals.
Propose a launch plan with success metrics, guardrails, and an experiment or evaluation design.

Constraints

You have two quarters for an MVP, with 1 PM, 1 designer, 6 engineers, 2 applied scientists, and shared support from trust and safety and data science.
You cannot train a new foundation model this cycle. You may use prompt changes, model routing, orchestration, retrieval, and evaluation on approved models.
The first launch must work in existing Grammarly Assistant and editing surfaces. A deeper flow inside Grammarly's document workspace can be phase 2, but you cannot require users to move writing into a new surface.
For legal and trust reasons, any substantive rewrite must be reviewable by the user before send or export.

Format

Lead with your recommendation and target segment. Then cover quality bar, scope, metrics, risks, and rollout in that order.