AI teammates

What is an
AI teammate?

An AI teammate is an AI agent that works alongside your team like a colleague. You hand it a task in plain words and it does the work itself: it uses your apps, browses the web on its own computer, remembers your business, works with other teammates, and checks with you before anything important.

Maya

Sales Manager · AI teammate

Finding leads · 4 steps · 38s

  • Its own computerlinkedin.com open
  • Your appsGmail, HubSpot, Slack
  • Company BrainKnows your customers
  • SkillsWeekly pipeline review
  • ScheduleFollow up tomorrow, 9:00
  • RulesAsks before sending

Needs you

Approve 20 emails?

Anatomy

What an AI teammate is made of

Eight parts turn a chatbot into a colleague. Every Octopus22 teammate has all of them.

01 · A role

A job you describe in plain words

Give it a name, a role and a few lines on what it owns. That's the whole setup.

  • Instructions you can edit any time
  • Yours alone, or shared with the whole team
  • Thinks harder when the task is harder, automatically
New teammate

Maya

Sales Manager

Working now

Instructions

You own our outbound pipeline. Find leads that match our best customers, write like me, keep HubSpot current. Never send without my OK.

Who can use Maya

Only meWhole team

02 · A computer

Its own computer, with a real browser

It doesn't just answer. It opens websites, clicks, types, runs code and makes files, on a cloud computer you can watch live.

  • Watch its screen live and take over with your mouse and keyboard
  • Real Chrome that stays signed in between jobs
  • Runs code and long jobs in the background
  • Makes Word, PowerPoint, Excel and PDF files
  • Hands you the screen for passwords, 2FA, captchas and payments
Maya's screen · live
linkedin.com/sales/search/people

LinkedIn wants a 2FA code.Maya handed you her screen.

Take over

03 · Tools

Every tool built in

Each teammate can research, create and communicate from day one. Nothing to install.

  • Web search, cited answers and deep research reports
  • Maps, places and directions
  • Creates and edits images, makes videos with sound
  • Speech from text, and transcripts from audio
  • Asks you with quick choice cards and short forms
Built in, on every teammate
Web search
Deep research
Maps
Images
Video
Speech
Transcripts
Code
Office files

Researching · 6 sources · 14s

04 · Apps

Works inside your apps

Connect the tools your business runs on and your teammate works in them as you, with your access.

  • 500+ apps, most in one click
  • Asks you to connect an app right in the chat when it needs one
  • Several accounts per app
  • Add your own MCP servers
Maya's connections
Gmail
Slack
HubSpot
Notion
Salesforce
Shopify
GitHub
Linear
Airtable
Canva
Figma
Asana
Plus any website, and your own MCP servers500+

05 · Memory

Remembers your business

What your team decides and prefers is learned from your conversations and kept in one shared Company Brain.

  • Every fact dated and linked to where it came from
  • Old values kept as history when things change
  • Correct or remove any fact in a click
  • Your personal preferences stay private to you
  • Searches past conversations when it needs to
Company Brain
Acme renewal owner

Priya

from #sales · Sep 12

Annual plan price

$49 per seat$39

from #pricing · Sep 20

How I like replies

Under 100 words, no jargon

from Chat with you · Sep 3

06 · Skills

Skills it learns once and reuses

Teach a teammate how your company does something, or let it write the skill itself after doing the job.

  • Write skills in plain words, or upload them with scripts and files
  • Give each teammate the skills its role needs
  • Teammates turn what they learned into new skills
Skill

weekly-pipeline-review

Written by Maya after last Monday’s review

  1. 1Pull every open deal from HubSpot
  2. 2Flag deals with no reply in 7 days
  3. 3Draft a nudge for each, in my voice
  4. 4Post the summary in #sales

Assigned to Maya and Noor

07 · Rules

Asks before anything important

Reading and research run freely. Anything that sends, pays, deletes or cancels waits for you, in apps and in the browser.

  • Every step judged on what it would really do
  • Approve or reject, with a note it reads
  • Per app and per action: ask when it matters, always allow, always ask, or off
  • Never asks for your password in chat
Needs your approval

Maya wants to send 34 emails

app Gmail

to 34 churned accounts

subject 3 months on us, if you come back

RejectApprove

Gmail · send email

Ask when it mattersAlways allowAlways askOff

08 · Schedule

Follows up on its own

A teammate sets its own reminders and recurring work, so nothing waits for you to remember.

  • Later today, at a set time, or on a repeating schedule
  • Repeats can stop on a date or after a number of runs
  • Runs in your company's timezone
  • See and cancel any of them
Maya's schedule

Check whether Acme replied

In 2 hours

Send the follow-ups you approved

Tomorrow, 9:00

Pipeline review, until Dec 31

Every Monday, 8:30

Set by Maya herself. You can cancel any of them.

One task, start to finish

How an AI teammate works

Follow Maya through one real job, from your first message to the follow-up she sets herself.

  1. 1

    Step 1

    You ask

    In a private chat, in a room with your team, or out loud. Plain words, the way you'd brief a colleague.

    You
    @Maya find 20 companies like Acme and draft an intro to each

    Typed, or said out loud with dictation

  2. 2

    Step 2

    It plans

    It decides how hard to think about the job and picks just the apps and tools it needs.

    Maya is thinking carefully

    Picked for this job

    HubSpot LinkedIn Gmail Web search
  3. 3

    Step 3

    It does the work

    On its own computer and in your apps. One live line shows progress, and you can open it to see every source, command and screenshot.

    Working · 6 steps · 1m 12s

    • Searched the web8 sources
    • Opened linkedin.com/saleson its computer
    • Built leads.xlsx20 rows
    • Drafted 20 introsin your voice
  4. 4

    Step 4

    It checks with you

    Anything that sends, pays or deletes waits in your Inbox for a yes or no. Sign-ins come to you as a screen handoff.

    Needs you

    Send 20 emails from Gmail?

    RejectApprove

    LinkedIn asked for a code, so Maya handed you her screen

  5. Step 5

    It delivers and remembers

    Files land in the Library, what it learned goes into the Company Brain, and it schedules its own follow-up.

    leads.xlsxSaved to Library

    Learned: “companies like Acme” means B2B SaaS, 50 to 200 people

    Checks for replies Friday, 9:00

Working together

How AI teammates work together

One teammate is useful. A team of them, splitting the work, passing it on and sharing one memory, is a department.

YouCole · booksIvy · receipts

01

Several take their part

One request can wake several teammates at once. Each takes the part that's its kind of work, and they run side by side.

e.g.“Close September”: Cole reconciles Stripe while Ivy chases missing receipts.

Juno · researchNoor · marketing

02

One hands off to the next

A teammate finishes its part and names who's next. Naming a teammate wakes it, with everything said so far.

e.g.Juno finishes the competitor brief: “Noor, it's ready for the comparison page.”

Juno

03

Big jobs fan out

A teammate splits a big job across background helpers that all run at once, then turns their reports into one answer.

e.g.Juno sends five helpers to read 40 competitor pricing pages in parallel.

Who can fix the invoice email?stays quietCole wakesstays quiet

04

The right teammate answers

There's no manager passing work around. Each message goes to the teammates it's meant for: a name always wins, otherwise the kind of work decides, and nobody replies to small talk.

e.g.“Who can fix the invoice email?” wakes Cole. Noor and Ari stay quiet.

Chief of staffNew: RecruiterYou

05

They hire a specialist

When work needs a role nobody has, a teammate can set up a new teammate for it. The new hire introduces itself to you.

e.g.Hiring piles up, so your chief of staff hires a Recruiter.

BRAINAri · supportCole · finance

06

One memory for all

What's decided in one conversation reaches every teammate a few minutes later, through the shared Company Brain.

e.g.Ari learns refunds over $200 need Priya's OK. Next week, Cole applies it.

Ivy · opsYouSamSamSupplier A?

07

People make the calls

Questions and approvals go to the people involved, and the first answer counts. It goes straight back to the teammate that asked.

e.g.Ivy asks which supplier to use. You pick one, and she carries on.

AI teammate vs chatbot vs automation

Same request, three very different outcomes.

You
Follow up with every customer who didn’t renew last month.

Chatbot

Answers questions

What it does

  1. 1Writes you an email template
  2. 2Explains how you could find the list

You still do the work.

  • Does the work itself
  • Uses your apps
  • Handles the unexpected
  • Remembers your business
  • Works while you're away
  • Asks before risky steps

Automation

Follows a fixed script

What it does

  1. 1Sends one template to a saved list
  2. 2Skips anyone the rule didn't expect

Works until something changes.

  • Does the work itself
  • Uses your apps
  • Handles the unexpected
  • Remembers your business
  • Works while you're away
  • Asks before risky steps

AI teammate

Owns the task

What it does

  1. 1Finds who didn't renew in Stripe
  2. 2Writes to each one about their account
  3. 3Asks you before anything is sent
  4. 4Checks for replies on Friday

It's done, and you approved it.

  • Does the work itself
  • Uses your apps
  • Handles the unexpected
  • Remembers your business
  • Works while you're away
  • Asks before risky steps

Ready when you are

Build the team
that does the work.