Hear the whole room
Most meetings hear from three people. Hivemind, a Google Meet add-on, hears from everyone, out loud, in a minute. I designed and built all of it, from research to working prototype.
Hivemind, in a minute
One question. Everyone answers out loud, at once. A minute later, the room hears itself.
- AskOne open question, from inside Google Meet.
- Answer out loudEveryone at once, on their phone. No names.
- Hear the roomA minute later, the room’s own words, played back.
- Run the meetingThe readout becomes the agenda.
People speak about three times faster than they type.
A poll gets what people bother to type. Hivemind gets what they think.
A self-initiated concept, tested with small groups of volunteers. The design system →
The brand
A name, a mark and two colours, settled before any screen.
It had to feel like a company, not a feature, from the first time it appears in a call.
Cobalt is the structure. Honey is the voice, and it’s kept scarce so it means something. The colour system →
In the call
It joins the call you’re already in.
A Google Meet add-on. The host opens it from Activities; everyone else scans a code. Nobody installs anything.
Built to what Meet allows: an add-on can’t touch microphones or chat, so the host mutes the room and Hive, a participant like a notetaker, does the speaking.
On your phone
One question, one tap, a minute of talking.
Talk for up to a minute. Trim words, but never rewrite them, and see the exact line that might be quoted before you send.
The room, heard
Ten answers become one readout, in people’s own words.
Ten minutes of answers, heard back in one: the headline, the themes, where it splits and the idea only one person had.
Quotes make the room hear itself. “Half the trials I talk to” says sales in a room of ten, so quotes are cleaned of anything that points to a person, and the speaker sees theirs first.
The rest of the meeting
The readout becomes the agenda.
The agenda is written after the room speaks, not before. Each item says why it’s there: “Raised by 6 of 10.”
Over time
What a team thinks, over weeks, and what to do about it.
One meeting tells a team what it thinks today. Twelve weeks tell it how it thinks, and what to do about it.
Meetings run on a few voices
Most rooms hear from three people.
In a meeting of ten, three talk. The other seven had views too; there just wasn’t a moment for them.
The fix is old: hear everyone first, without names. It takes a facilitator and twenty minutes. Hivemind takes one.
A few voices carry the room
Most of the talking comes from a few people. The rest is said in the corridor after.
Meeting research; widely observed
The first and loudest set the frame
Whoever speaks first, or most senior, sets the direction.
Anchoring and HiPPO literature; Nominal Group Technique (PMC 4909789)
People go along with what they assume others think
Everyone privately disagrees, and assumes they’re alone.
Pluralistic ignorance; The Decision Lab
Independent answers first works
Hearing everyone first, without names, surfaces more ideas. It just takes too long.
Nominal Group Technique; Delphi method
What already exists
Every piece exists. The loop doesn’t.
If Microsoft or Google already did it, there was no product. So I checked first.
Mentimeter
Slido
AhaSlides
Live polls with AI grouping
Collect typed answers from a room and group them into themes.
MissesTyped, so short. A poll result on a slide, not a step in the meeting.
Remesh
ThoughtExchange
Large-group research
Gather and synthesise open answers from hundreds of people.
MissesBuilt for researchers and town halls, run outside the meeting. Typed.
Teams Facilitator
Gemini in Meet
Zoom AI Companion
Meeting assistants
Notes, recaps and agendas, built into the call.
MissesThey summarise who spoke. The people who stayed quiet are missing from the record.
Otter Meeting Agent
Voice agents in the call
A voice in the meeting that answers questions.
MissesIt answers one person. It doesn’t hear from everyone.
Cloverpop
Decision logs
Record decisions and the reasoning behind them.
MissesAfter the fact, typed up by one person.
- Teams polls
Meet polls
Built-in polls
Quick anonymous votes and Q&A, free in the call.
MissesClosed questions. They count, but don’t listen.
| Category | Everyone answers | Out loud | Turned into themes | Played back to the room | Runs the meeting |
|---|---|---|---|---|---|
| Live polls with AI grouping | Yes | No | Yes | Partly | No |
| Large-group research | Yes | No | Yes | No | No |
| Meeting assistants | No | Yes | Yes | No | Partly |
| Voice agents in the call | No | Yes | Partly | Partly | No |
| Decision logs | No | No | No | No | No |
| Built-in polls | Yes | No | No | Partly | No |
| Hivemind | Yes | Yes | Yes | Yes | Yes |
Notetakers record whoever spoke. Polls collect what people bothered to type. Hivemind lets everyone talk at once, and turns all of it into the next thirty minutes.
The risk: Microsoft or Google add “ask everyone” and give it away. The answer is the whole loop, and the history it builds.
Turning it into a business
Free in the meeting. Paid for what builds up.
The meeting is the marketing. Every round shows Hivemind to a whole team at once, and only the host signs up. The launch site →
A host tries it
Installs it and asks one question. Everyone else joins with a code.
FreeThe room hears itself
The moment that sells it, to ten people at once.
FreeIt becomes a habit
Planning, retros and kick-offs open with a round.
TeamThe picture builds
The dashboard shows how the team thinks, and what to try.
TeamLeadership looks across
What teams keep raising, aggregated.
Org
Where it converts. The free plan’s limit is set just under a weekly habit, so the upgrade arrives the week a team has proved it wants this.
- 100InstallFrom the Workspace Marketplace, or a readout link someone was sent.
- 60Run a first roundActivation. The first readout is the product’s best salesperson.
- 30Hit the Free limitFour rounds in a month: a weekly habit. The upgrade prompt appears here.
- 8Pay for TeamAbout 1 in 4 habitual teams.
- Free
$0
Three rounds a month, the readout and the agenda.
- Team
~$10per seat a month
Unlimited rounds, the dashboard, suggestions and history.
- Org
Talk to us
Cross-team insights, SSO, data controls.
What a team is worth. Priced by the seat, costed by the round.
$960 a year, per team
Eight seats at $10 a month: the size of a team that meets weekly.
About 10c a round to run
Ten minutes of speech at $0.006 a minute, plus the summary. Roughly $2 a team a month.
~95% gross margin
The cost is per round, the price is per seat. Margin grows with the habit.
How it grows. No sales team until Org.
Every round is a demo
Ten people see it work. The readout page carries “Run your own”.
Free stays generous
Three rounds a month is enough to prove it, not enough to live on.
Leaders buy across teams
Org is sold once three or more teams in a company pay.
The first twelve months. Each phase has one number that says whether to go on.
- Months 0–3
Prove it
Run the three tests. Build the Meet add-on and real speech-to-text. Private beta with ten teams.
Readout judged faithful by 8 of 10 teams
- Months 3–6
Launch free
List on the Workspace Marketplace. Measure activation and the weekly habit, and fix what drops people.
60% of installs run a first round
- Months 6–12
Charge for the history
Turn on Team: the dashboard, suggestions, Slack and calendar. Start Org conversations.
200 paying teams, ~$16k a month
All figures are proposals and assumptions; nothing here has been sold. Speech-to-text at a published rate of $0.006 a minute.
Rules before screens
The words and the rules came first.
The words came before the screens: one name for each thing, everywhere.
- Hivenot session
- Roundnot poll
- Voicenot response
- Readoutnot results
- Themenot cluster
- Quotenot verbatim
- Agendanot action items
- Live contextnot notes
No purple. It’s the colour of every AI product. Honey is owned by nobody in meeting tools. Five design passes →
Safe to say it
Nobody can tell who said what. Every screen is built on that.
Hiding names took one decision. Everything else that gives a person away became a rule, not a setting.
Arrival order
Dots that light in order show who was quick.
The ruleFill by count, never by person.
Room size
In a room of three, every answer is traceable.
The ruleNo readout under four voices.
Quotes
“The trials I talk to” says sales.
The ruleCleaned, and shown to the speaker first.
The laptop mic
It’s usually the call’s microphone.
The ruleMute the room; answer on phones.
Transcripts
They outlive the moment.
The ruleOnly the speaker sees theirs. Audio is deleted.
Editing after
People would tune answers to the room.
The ruleTrim before sending, never after.
Testing it honestly
Each belief got a test, with the pass mark written before it ran.
A concept is easy to make look finished, so I ran every test with small groups of volunteers. Small groups show direction, not rates, and each one changed the design.
- H-01
A few people carry most meetings, so most of a room’s thinking is never said out loud.
- H-02
Speaking an answer is faster and fuller than typing one, so people say more of what they think.
- H-03
Hearing its own words back changes what a room talks about next.
- H-04
People only speak freely if nobody can work out who said what, quotes included.
- DoneChecks H-02
Speak or type
A small group of volunteers, one open question each way
Do people say more, and more of what they think, when they talk?
Passes ifSpoken answers carry at least twice the distinct points of typed ones, and 4 of 6 prefer speaking.
What happenedSpoken answers ran longer and carried more distinct points than typed ones.
Most preferred talking. A couple didn’t want to talk out loud in an open-plan office.
Changed: Typing stays as a quiet fallback on the phone.
- DoneChecks H-01, H-03
One real round, then the agenda
A volunteer group, in a real planning conversation
Does the readout change what the meeting talks about, and does the room follow it?
Passes ifThe meeting works through at least two agenda items from the readout, and most say they heard a point they would not otherwise have heard.
What happenedThe group worked through the readout’s agenda, and the point most hadn’t heard before took the most time.
People wanted to know why each item was there.
Changed: Every agenda item now says why it’s there: “Raised by 6 of 10.”
- DoneChecks H-04
Can you tell who said it?
Volunteers who know each other well
Do cleaned, anonymous quotes keep people anonymous among colleagues who know each other?
Passes ifGuesses are right no more often than chance.
What happenedMost guesses were wrong. The ones people got right had a giveaway phrase, like a team or a role.
Changed: Quotes are cleaned of anything that points to a person, and the speaker sees theirs first.
What I’d do next
Test it with real teams, then build it for real.
Write the rules before the screens
They made most of the later decisions for me.
Prototype the system, not the pictures
One simulation found what no mock-up would: dots that showed who answered first.
Next: test with real teams over a few weeks, build the real add-on with speech-to-text, and check the readout stays faithful to what people said. Run the prototype yourself →