The voice AI company ElevenLabs previewed a work app called ElevenAI at its New York summit and opened free early access the same day, by signup rather than as a normal download. In meetings, including in-person ones, it takes notes after letting everyone know it is recording, then gives each attendee a summary and their own action items. It also turns speech into cleaned-up text in any app, in more than 90 languages, and a company can connect its AI agents so they act on the notes, for example by drafting an email.
The notes themselves are no longer rare
Wispr Flow added a notetaker to its dictation app in September. Granola lets teams share notes and ask questions across them, and Fireflies does similar work. So putting dictation and meeting notes in one app is not what is new here. What ElevenLabs changes is who the app is for: a whole company, with its agents attached, instead of one person's notebook.
When several products write good enough notes, the notes stop being what sets them apart. A team that pays a separate fee for each person to get transcripts is paying for the most common part of the job.
The typing pool shows where the work goes
The dictation machine tells the same story. Even after decades of marketing, few managers used it themselves; it was a tool for typing pools and private secretaries. Better recorders did not end the pools. Word processing did, and then the personal computer in the 1980s put a keyboard on every desk, so each person became their own typist.
A notetaker does that to minutes. Every attendee, not only the executive with an assistant, now gets a written record of the meeting. The step after that is one the pools never had: software that also drafts the email, so the person who spoke in the meeting is not the one who has to write it up afterwards.
What it gives back to one person
Take a recruiting manager who runs ten meetings a week. After each one she spends about five minutes writing who will do what and sending it round. That is 50 minutes a week, roughly 40 hours over a working year, a full week of her time. If a draft is waiting when the call ends and she needs one minute to fix it, she gets back about 30 hours a year.
That minute is the part that stays with her, because a transcript hears words and not intent. When someone says "let's try Thursday", a draft can turn it into "confirmed for Thursday". The person who knows it was only a maybe is the one reading the draft.
When a team compares these apps, the useful test is one real meeting: which app hands back a follow-up you would send after a single fix.