Why review-first AI matters for email and message workflows

A missed commitment can matter, but over-tracking every casual phrase creates fatigue. Vortyx is designed around a review-first principle: AI can surface likely work, but users should stay in control of what becomes tracked, edited, dismissed, or sent.

Vortyx can surface follow-ups, commitments, waiting items, next steps, recurring responsibilities, draft replies, draft follow-ups, and scheduling suggestions. These are suggestions. They should remain reviewable until the user decides what to do.

Can you send the updated proposal next week?

Vortyx may suggest a follow-up or commitment from this message. The user can confirm it, adjust the date, change the title, or dismiss it.

For email drafts, Vortyx may suggest reply or follow-up text. The user can edit the draft, improve it with guidance, or discard it. Vortyx does not treat generated text as approved communication.

Automatic execution is risky in communication workflows because intent is often ambiguous.

Let's reconnect after launch.

This could be a real follow-up, a polite closing, or a vague idea.

I'll try to send it tomorrow.

This suggests intent, but the commitment may be uncertain.

Can someone check this?

The owner may be unclear. In these cases, full automation would either miss useful work or create false positives. Review-first design allows Vortyx to surface possibilities without forcing them into the user's workflow.

• Suggested items appear for review before becoming tracked work.
• Draft replies and follow-ups remain editable.
• Email actions are not sent without user approval.
• Users can dismiss suggestions that are not useful.
• Users can snooze or mark items handled.
• Recurring commitments require explicit setup or confirmation.
• Needs Review separates proposed work from confirmed tracking.
• Needs Attention surfaces tracked items that may require action, rather than silently acting on them.

These safeguards matter because they make false positives manageable.

User review behavior is also useful feedback. Vortyx can evaluate which suggestions are accepted, edited, dismissed, snoozed, improved, or marked handled.

These signals can help improve detection quality over time, but they should not be presented as accuracy metrics without a controlled evaluation method.

Vortyx is currently evaluating these signals with a closed set of users. Quantitative metrics will be published only after there is sufficient repeatable evidence, including a labeled evaluation set, annotation rules, sample size, precision, recall, false-positive rate, and model/version context.

Vortyx should be clear about its limits. It does not automatically send emails. It does not approve draft replies. It does not silently create external commitments. It does not assume that every future-looking statement is a task. It does not replace the user's judgment.

The product promise is not that AI does your work for you. The promise is that important work signals become visible and reviewable before they slip away.