Superhuman’s new auto-draft feature almost makes me like AI replies

Superhuman is rolling out an upgraded AI auto-draft feature for its email client, improving relevance and tone of replies and enabling users to send some drafts with little or no edits. In beta, co-founder Rahul Vohra says 40% of auto-generated drafts were sent within one day and 60% of those were sent without manual editing. The system uses a mix of frontier models from Anthropic and OpenAI to leverage more context, with the company also advancing toward a cross-platform assistant (“Superhuman Go”).

Analysis

This is a quiet but important signal that AI is moving from a feature to a baseline expectation in white-collar workflows. That shifts value away from the email client/UI layer and toward whoever owns identity, context, and cross-app distribution; in public markets that favors MSFT and GOOGL more than standalone productivity apps. The second-order effect is pricing pressure: once users expect decent drafts everywhere, point solutions must justify a premium on trust, workflow integration, or compliance rather than on raw model quality.

Near term, the adoption curve matters more than the product demo. If usage data shows high send-through with low edits, it supports seat expansion and lower churn for premium collaboration suites over the next 1-3 months; if not, these features remain retention sugar rather than monetization. Over 6-18 months, the key risk is inference cost: if assistant-heavy products cannot pass through model spend, gross margins compress even if engagement improves.

Contrarian view: the market may overestimate how quickly AI drafting reduces labor or transforms productivity economics. Faster email throughput can also increase message volume, which benefits filtering, governance, and enterprise controls more than end-user subscriptions. The thesis is falsified if enterprise buyers refuse AI-generated outbound on policy grounds, or if model costs and latency keep forcing manual edits despite polished demos.

Key Decisions for Investors

• Long MSFT / short WCLD for 3-6 months: express the view that AI becomes a bundled feature in platform software while weaker SaaS names face feature commoditization and margin pressure.
• Overweight GOOGL vs. broad software beta: the firms with native model stacks and distribution should capture more of the workflow attach rate than wrapper apps; use a 1-3 month horizon into enterprise AI budgeting updates.
• Buy a modest MSFT or GOOGL call spread 90-180 days out if you want convexity to AI-assistant monetization improving retention, but keep size small because this is a feature adoption story, not a near-term revenue re-rate.
• Watchlist, not trade: if model-inference costs are disclosed or implied by next-quarter gross margin commentary at AI-heavy SaaS names, that will determine whether this is a durable margin tailwind or just engagement noise.
• Avoid chasing standalone AI productivity-app momentum until there is evidence of pricing power; the falsifier is sustained willingness to pay above incumbent suite pricing for assistant-only functionality.