Own your intelligence

AI teammates that own the outcome.

Give an AI teammate an outcome and the KPI that measures it. It learns how your business works from your systems and from how your people did the work, is tested on your past cases, then delivers. Your people approve each step, and what it learns stays with your company.

Mac-only for now. Free during the beta (invite-only).

On Jamie's Overview, under the outcome Refund requests answered the same day, inside policy, Okafor Freight writes that order 4417 arrived damaged and asks for a refund, Jamie checks Refund policy v2, where the line Damaged on arrival: full refund up to $500 is marked, replies that the $320 has been refunded to the card and will show in 3 to 5 days, and the request is marked Answered in 4 min, then a $1,200 refund request from Halvorsen Build is over $500 so Jamie's draft waits for Sam Rivera, who presses Approve and send, and the KPI Answered the same day rolls from 87% to 88% beside Your team before: 64% and Target: 90%.
On the Tests page of Jamie, an AI teammate, the three sources of how your people did it are read, the Support desk with 1,240 tickets your team answered, Company memory with Refund policy v2, approved by Priya, and Customer records with 214 accounts, then one past case, order 3958 that arrived damaged, sets what Sam Rivera did, refunded in full and cited Refund policy v2, beside what Jamie would send, a refund in full citing the same policy, and is marked Matches, then all 240 past refund requests run through until 212 of 240 match how your team answered and 28 differ, for a person to check, with Review 28 beside them, and the page names the people it learned from, Sam Rivera, Nina Park and Maya Chen.

Who is Them?

You already know Them.

sees problems differently from everyone else
has years of judgment your company depends on
knows how things work beyond the process documents
remembers why important decisions were made
has expertise that would take years to replace
knows what to do when there is no playbook

Them is who everyone turns to when the obvious answer is not good enough.
People and AI teammates like Them help build companies. Them AI helps their expertise keep building.
Every AI teammate is measured against how Them did the work.

Where Them AI fits

Assistants answer. Agents finish tasks. AI teammates own the outcome.

AI assistants

One chat, one person

It replies to what you ask, and what it learns stays in that chat.

AI agents

One task, then it stops

It does the job you hand it. The next one starts from zero.

AI teammates on Them AI

Held to your number

You give it an outcome and the KPI that measures it. It learns how your business works, starts the work when something changes, and is measured against how your people did it. You decide what needs your approval.

Put an AI teammate to work.

Four steps, from an outcome in plain words to work it starts in your team's room.

  1. Tell it the outcome you want.

    Name the outcome, like every renewal going out on the terms you approved, and the KPI that measures it: a number your team already tracks, like the share of renewals sent on those terms. A few short questions set its job, its limits, and what waits for a person.

  2. It learns how your business works.

    It reads the systems you connect and how your people handled the work before: past requests and renewals, and what your people decided on each. It keeps its own notes, learns more each week, and asks about what it cannot know. Company memory stays what your people approved.

  3. Tested on your own past work.

    It is run on your past cases and scored against what your people did. People who build AI call these tests evals. Tell it what was wrong in plain words, and each correction becomes one more test its work has to pass.

  4. It delivers in the room with your team.

    It picks up requests in its room, the shared space where your people and AI teammates work, and starts work on its own when something changes: a renewal date coming up, a new ticket. It writes in the room, does the work and asks the person it is for to approve what waits for them.

    Coming next: Slack and Microsoft Teams, so the conversation and the approvals can happen there too.

In the setup chat for a new AI teammate, Them AI asks what outcome it should own, Maya Chen answers that every renewal goes out on the terms the team approved, the chip KPI: share of renewals sent on approved terms lands under her answer, the steps tick one by one, each after its answer, Outcome, KPI, its job to prepare each renewal and its note, what it works on, renewals in the CRM, and Waits for a person, she picks Anything a customer will see for what waits for a person, then she presses Create Sage and get it ready and Sage, the new AI teammate, appears beside the button.
Sage gets ready: three sources, Company memory with 42 approved records, a CRM with 214 accounts and a Support desk with 1,240 tickets, flow into what it has now, each source's tick adding to what it has now, up to 545 notes it wrote down while reading, 38 past cases, renewals your people decided, to test it on, 9 drafted tests made from past cases and 4 questions about what only your team knows, and one note, Halvorsen Build renews in 60 days, is shown close up, marked as Sage's note that no person has checked yet.
Sage is tested on past cases, 31 of 38 of which match what the team did, and on one where the team held the discount at 15 percent and asked Priya while Sage would have offered 18 percent, so Maya Chen presses Correct this and types never go past 15 percent without Priya, and Sage answers that it added that as a test.
In the Pricing desk room a renewal date is coming up, so Sage writes in the room, telling Sam Rivera that his brief for the Halvorsen Build renewal is ready with a draft note to their finance lead, and attaches two files, the brief and the draft note, which waits for Sam Rivera with Approve and send.

The first four weeks

It earns each step, and a person decides every one.

  1. Week 1

    People approve.

    Every draft waits for the person who asked. Nothing goes out until they send it.

  2. Week 2

    It starts the work.

    It picks up requests in its room without anyone asking it by name, and starts work when something changes. Its drafts still wait for a person.

  3. Week 3

    It posts on its own, once earned.

    Once its drafts keep going out unchanged, you are asked whether it may post on its own. Whatever you chose to keep for a person still waits.

  4. Week 4

    Its workflows run on their own.

    Work it repeats is saved as a workflow that runs start to finish, and stops only at the approvals you kept.

The weeks are a guide, not a timer.

Start from the outcome you want.

Ready-made AI teammates and skills, free to add during the beta.

Renewal desk · AI teammate

Every renewal checked against the terms you approved.

Reads each renewal against your renewal terms and discount policy, and lists what needs a decision before anyone signs.

Support refund desk · AI teammate

Refund replies that follow your policy.

Sorts refund and credit requests by your refund policy, works out the amount, and drafts each reply for a person to send.

Proposal drafter · AI teammate

First drafts with every price cited.

Turns your discovery call notes and approved pricing into a first draft proposal, with every gap marked.

Discount policy check · Skill

Discounts checked against your policy, with the approver named.

Checks a proposed discount against your discount policy, and says who has to approve it.

Meeting notes to decisions · Skill

Decisions that outlast the meeting.

Turns raw notes into decisions, owners and dates, and suggests which ones are worth keeping as company records.

Your own · AI teammate

An outcome of your own.

Build one around an outcome you care about, from your own methods or from the approved work of the person who knows the job best.

Next, each AI teammate will start the same way: an outcome and its KPI, tests on your past cases, then the first four weeks.

How it stays in your hands

What makes an AI teammate safe to hand an outcome to.

Notes, and company memory

An AI teammate keeps its own notes as it works. Each one is marked not reviewed, and you can open them all. Company memory is only what your people approved, cited in every answer that uses it.

You decide what waits

Anything a customer will see, only prices and contract terms, or nothing at all, with its work reviewed afterward. Every decision is on the record with who made it.

Tests on your past work

Each correction becomes a test. A later change to the teammate has to pass every earlier test before a person approves it, so a fix does not break what already worked.

Measured against your people

Its KPI is worked out from the systems you connect and shown beside how your people did before it started. The number is for you to read. A person still decides what the teammate may do on its own.

Work it starts is on the record

Every piece of work an AI teammate starts is listed on its page with what started it. Anyone in the room can stop it, and it stops at the approvals you keep.

Your model, your memory

Run on the AI model you choose. The methods your team approves export as plain files, and approved memory reaches the other AI tools you connect with a key.

Who it is for

Does your best answer usually start with "Ask so-and-so"?

Them AI is built for the person who answers for an outcome, and the team whose work depends on what the company already knows.

Support and customer success. Sales. Consulting and advisory. Operations.
If a few people hold the context everyone needs, or useful AI work stays in separate chats, bring that thinking into one memory your whole company owns.
We onboard teams in small groups, with a person helping each one get started.

How to start

Got a team in mind? Bring them.

Join the waitlist with your business email.

Bringing your whole team? Email hello@usethem.ai with your team's size and what you work on. Teams joining together get priority.

Already know someone using Them AI? Ask them for an invite from their workspace.

Available on Mac. Free during the invite-only beta.

Contribute individually. Build collectively.