Breaking

Superhuman Rolls Out Revamped Auto Drafts, Blending Anthropic and OpenAI Models for Tone-Matched Email Replies

July 14, 2026 at 10:04 EDT

AI Agents
Sales & Business
Foundation Models

Superhuman, the premium email client now owned by Grammarly, is beta-testing a significantly upgraded version of its Auto Drafts feature that generates full reply drafts in a user's own tone—an update that early testers say is finally good enough to send with little or no editing.

Superhuman · AI Auto Drafts

AI email replies that finally sound like you

Superhuman's revamped Auto Drafts writes full replies in a user's own voice — pulling from their writing history and personalization settings. Early testers say the drafts are now good enough to send with little or no editing.

~40%

of auto-generated drafts are sent within a day

~60%

of those go out with no manual edits at all

reply variations offered per message

Of the drafts users actually send…

Each block = 10% of sent drafts

60%

sent with zero edits

40%

lightly edited first

What changed under the hood

MODELS

Frontier models from Anthropic + OpenAI, replacing the old GPT-3.5 Instant Replies

CONTEXT

Larger context window pulls from your history + personalization settings

LEARNING

Adapts to feedback — e.g. stops suggesting late-night meeting slots once corrected

What works

Matches your tone well enough to send as-is
Handles routine replies: embargoes, meeting times, declines
Fits the fast, keyboard-driven workflow across Gmail & Outlook

Still rough

Drafts skew positive — too readily agree to pitches
Can still propose inappropriate meeting times
Free Gmail & Outlook AI overlap; a human check still required before sending

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.