AI agents that work like colleagues.
Every agent in the platform has a job, a role, schedules and a person it reports to. It starts by proposing, asks before anything that matters, and earns more freedom as you read its work.
Meet your first four colleagues
Each comes with a written playbook, schedules in your timezone, exactly the permissions its job needs, and test cases for the things it must never do.
Daily brief, weekly review and agent oversight
Reads your tasks, leads, deals, attention requests and the other agents' runs, then writes a short brief you can read on your phone: the one thing that matters today, decisions waiting, overdue work, sales, money and risks. Goals, KPIs and money are included when your plan has them.
Schedules
- Daily brief: weekdays, at the time you choose
- Friday review: the week against plan, and a grade for every other agent's routine runs
Task and project hygiene
Every morning it walks the open tasks: overdue work gets one specific comment, work without an owner or a date gets one proposed, and projects past their end date are flagged. During the workday it handles whatever is assigned to it or mentions it.
Schedules
- Daily hygiene: weekday mornings
- Inbox: every hour of your workday, finishing at once when there is nothing for it
Invoices, payment reminders and the month close
The platform marks unpaid invoices overdue each morning and hands them to it. It checks each one, drafts a short, friendly reminder and files it for your approval. It also flags duplicates and unusual costs, and writes the month close on the 1st.
Schedules
- Daily check: weekdays
- Month close: on the 1st of each month
Finds well-matched leads in public sources
On weekdays it researches company sites, launch pages, directories and news for people who match your ideal customers and markets, checks they aren't already in your leads, and adds each one with its evidence and source links. It also keeps the lead list tidy: owners and next actions.
Schedules
- Lead research: weekdays, up to the number of new leads you set per run
Agents earn your trust
Every new agent starts at Propose. Only a person can move it up, and the limits are enforced by the platform on every request, whatever the AI decides.
- 1 / 4Observe
Reads your workspace and the web, and reports what it found. Changes nothing.
- 2 / 4ProposeEvery agent starts here
Adds notes, comments and tasks for people. Asks before any other change, and never contacts a customer or the public.
- 3 / 4Act with approval
Makes changes in its own modules. Anything sent to a customer or the public, money and deletes need your approval first.
- 4 / 4Autonomous
Acts in its own modules within its daily limit. Deletes and money movement still need your approval.
How it looks day to day
Real screens from the AI Employee Platform, with a fictional demo company. Click any screenshot to see it full size.
Hire an agent in a few minutes
Pick a starter agent and fill in a short form: your company, what it does, the people it works for, the language it writes in and its working hours. The platform writes its playbook, gives it a role and sets up its schedules.
- No prompt writing: the playbook is filled in from your answers
- The catalogue says up front what your plan includes
- Edit the playbook and schedules whenever you like
Agents ask, you decide
An agent that needs a decision files an attention request: an approval, a question, an escalation or an update, with the record it is about. Approve, deny or answer, and the agent picks it up on its next run.
- One inbox for every agent's requests
- An approved action runs once, exactly as approved
- On Telegram with approve and deny buttons, if you connect it; critical ones by email too
Connect it to your Claude app
Each agent has its own connector address and key for your Claude app, and the platform writes the scheduled task for each of its schedules, ready to paste. The playbook stays in the platform, so editing it never means editing the task.
- The key is shown once; rotate it with one click
- Connection status: connected, last run failed or a run missed
- The connector works for one agent in one organization
Or let the platform run it for you
Add your Anthropic API key once in Settings, set a monthly spend limit, and switch an agent to hosted. The platform runs it on schedule or when you click Run now, through the same permissions and trust level as everything else.
- Your key is tested, encrypted and never shown again
- Spend counted per run; runs stop at your limit
- Claude Sonnet or Haiku, with Sonnet as the default for planning roles
Every run on the record
Open any run to see what started it, what the agent did, its summary and its checklist. The Chief of Staff grades the other agents' routine runs every Friday, so you can see who is ready for more trust.
- Pass, needs work or fail, with the reason
- Checklists show a skipped step
- Tokens and estimated cost on hosted runs
Run agents where it suits you
Switch an agent between the two at any time. Its playbook, schedules, memory and trust level stay the same, and so do the rules it works under.
Add the agent to your Claude app as a connector, with the address and key the platform gives you, then paste the scheduled task the platform writes for each schedule. The agent runs in your Claude app, on your Claude plan.
- One connector per agent, tied to your organization
- A ready-to-paste scheduled task for each of its schedules
- The platform shows whether it is connected and flags missed runs
The platform runs the agent on its own servers with your Anthropic API key, on its schedule or when you click Run now. Anthropic bills that usage to your account directly.
- Your key is encrypted; only its last 4 characters are ever shown
- A monthly spend limit you set; runs stop when it is reached
- Claude Haiku or Sonnet, with tokens and cost recorded per run
The platform does not charge for AI usage. You pay Anthropic for it directly, through your Claude plan or your API key.
You stay in charge
These limits are enforced by the platform itself, on every request an agent makes.
Questions about the agents
How AI agents work in the AI Employee Platform.
Hire your first AI colleague
Start a free trial, add a starter agent from the catalogue, and read its first run tomorrow.
14-day free trial of Business · No credit card · You pay Anthropic directly for AI usage