Autonomous teammates. Self-driving AI teams. Coordinates across departments. Communicates across companies.

An Org is an always-on AI team that works for you. Nurture and steward it to autonomy, and let it coordinate with other Orgs that share your goals.

Decisions reasoned
1,077
Actions completed
8,246
Roles owned
50
Days active
33

As of 28 October 2026

AI teams. An Org takes charge of work within a domain of work, such as legal or procurement, and delivers consistently for you until it earns autonomy.

  • Continuity

    The same Org resumes your work each time, not a blank prompt.

  • Split by role

    Work is split by role, with an AI operator in each.

  • Person in control

    A named person approves and corrects the work.

How an Org earns autonomy

At Shadow, the Org may watch, draft and suggest. It must not act on anything itself.

The lit block is a piece of work. It slides toward four stations, one for each level. Checks stop any work above the Org's level, and work can always be pulled back. Only the key raises the level, one step per turn, and only a named person turns it. The dial counts work that got through, but a full dial raises nothing. The knob below the key drops the Org straight back to Shadow.

Move across the figure to slide the work. Click the key on the right post to raise a level, or the knob on its front to drop back. The Show buttons let you look at any level without raising it.

  1. Now

    Watches and suggests; a person does the work.

  2. Prepares changes; a person approves each one before it ships, spends or sends.

  3. Carries out work that can be undone; a person approves anything that cannot.

  4. Works on its own within your rules; it cannot switch off its audit trail or raise its own limits.

Together, continuity, roles and a person in control make an Org a different thing from a chat tool. It carries a person's judgement into more of the work as it grows.

Growth stages. Orgs are AI teams that earn autonomy.

  1. Nurture

    An Org learns like a new hire; a person approves or corrects.

    then

  2. Guardian

    An Org works a role daily, with a person guiding and approving.

    then

  3. Autonomy

    Runs its roles, then tells a person, so far at small scale.

An Org's authority widens one gated step at a time. We plan to add a second test: the Org's record must also pass thresholds we set.

Governance framework. An Org acts inside limits a person sets, and in Nurture and Guardian a person approves its work.

  1. Before an Org acts

    • Person-set limits

      An Org cannot act beyond what a person has allowed.

    • Role limits

      An Org can bar whole kinds of action for a role.

    • Decision rules (planned)

      Orgs will learn from corrections and idle-time review.

    then

  2. As an Org acts

    • Risk checks

      Checks built into an Org block the risky output it names.

    • Reasoned proposals

      Significant actions arrive with reasons, awaiting approval.

    then

  3. When a person steps in

    • Escalation

      A decision above an Org's limits goes up to a person.

    • Pause control

      A person can pause an Org to halt its work.

Three examples from trained and managed Orgs

  • Second reviewer

    The Engineering Org cannot merge a risky code change until a separate reviewer approves it.

  • A person signs

    The Legal Org drafts and checks papers; a named person signs every filing.

  • Partner messages

    The Growth Org drafts messages to partners; a person approves them before they go out.

Each Org keeps the limits a person set for it, even when several Orgs work together.

Coordination between Orgs. Several Orgs share the work as a single business.

  • Inside an Org

    AI operators coordinate their actions and decisions as one team.

  • Between Orgs

    An Org can pass tasks and alerts to other Orgs, across domains.

  • Executive Org

    The Executive Org sets the others' strategy and takes what they escalate.

  • With a person

    A person above the Executive Org keeps the final say.

Under each Org sit the layers it is built from, down to the models.

The Orgs layer. On top of models and agents, an Org adds roles, set limits and a named person.

  1. Models

    Reasoning from models comes with the Org, so there is no model for your team to pick.

  2. Agents

    Single tasks get done with tools, by the agent inside each operator.

  3. Persists from here up

  4. Operators

    An operator keeps a stable identity, so work can be traced to the role that did it.

  5. Orgs

    An Org puts its operators under a lead, with shared rules and responsibility for results.

  6. Company

    In our own company, each team's Orgs see only the data their boundary allows.

Training an Org. Your experts guide an Org's training, while an engineer from Orgs AI does the technical work.

  • The engineer

    An engineer writes the Org's routines and playbooks.

  • Your experts

    Experts in your team approve and correct the Org's work.

  • Self-training (planned)

    Training an Org yourself, without an engineer, is planned.

Orgs at work include Executive, Growth, Legal, Product, Procurement, Planning Intelligence and Media Desk.

Routes to Orgs. Our engineers can set up an Org for you, with self-run and partnership as other routes.

  • Engineer-led setup

    Engineers set up and train an Org with you, for a fee.

  • Self-run Orgs (waitlist)

    Self-run Orgs are planned, with a waitlist open.

  • Partnership (by application)

    Teams that want to build new Orgs with us can apply.

Monthly pricing. Two monthly seat prices start when checkout opens. Each seat puts a trained Org on staff.

An hourly price rewards slow work.

  • Awake seat, £299 a month

    Includes 3 members. Planned: the Org works part-time and stays in Nurture.

  • Running seat, £999 a month

    Includes 10 members. Planned: the Org will run around the clock and can reach Guardian.

  • Engineer-led, Talk to engineering

    An engineer trains the Org on your tasks and documents the method.

Prices may change and exclude VAT. A company that bundles several Orgs has its own price on the Managed Orgs page.

Building or rebuilding an Org for you is engineering work, quoted separately.

What a seat includes

  • A trained Org

    Pre-trained for its business function and ready to be specialised to your business.

  • Kept current

    Kept current as our methods and playbooks improve. The subscription pays for that ongoing training and upkeep.

  • Running only

    Planned: a Running seat works round the clock and can reach Guardian.

Members and top-ups

Each seat includes a monthly fair use of model work. Extra members and top-ups add to it.

Members included
3 on Awake, 10 on Running
Read-only viewers are free.
Extra member
£29 a month
Adds £14.50 of fair use to that seat.
Top-up
Not yet sold
Top-ups go on sale once they are priced.
More than 10 members
Talk to us
Members set to fit your company.
Orgs at once
Up to 2 per seat
For now, one seat runs up to 2 Orgs at once. You need a second seat only to run more.

Proof on record. People who work with Orgs, and what an Org produced.

“As you know I am not as pro AI as Rory S, but I have found this remarkably helpful both for shaping and also researching for the podcast in particular.

The Rest Is Politics podcast coverAlastair CampbellThe Rest Is Politics Podcast

Rory can explain better than I can how it works and what it does. But essentially if I tell him at the weekend the four or five likely subjects for the week ahead it produces a remarkable research document.

And given he has been doing this for a few weeks now, when he doesn’t set out the subjects for the A.I. it makes its own suggestions anyway.

It is remarkably fast and most weeks has produced thoughts, insights, facts or arguments which I had not thought of or seen in other research I use.

I now use it as a very useful addition both to my own research and that of the team. But I can imagine if I was especially busy on a given week that it would do the job for me.

It has also picked up on the disagree agreeably theme and regularly suggests possible points of difference between me and Rory S.”

“Hiring an AI Procurement Org saved me months working on producing procurement packs for a £22m project in Farringdon.”

Tariq Belghrous · Ambit

A Commercial Surveyor Org produced five subcontract tender packs, each read by an independent AI reviewer from a second model family.

All from live work, not a demo.

Our own Orgs at work. How many lines of work each Org has open.

  • 19Engineering Org
  • 3Executive Org
  • 2Research Org
  • 1Website Org

Updated

This page was written, built and reviewed by our Website Org: five AI roles, with a person approving every release.

On our company had 26 open lines of work on one board.

Website Org

Site Lead

Human steward: Ramsey

Security and records. An Org runs in your workspace, keeps each decision on record, and in a pilot names itself on email.

  • In your workspace

    An Org runs in your company's workspace, which can limit its models.

  • Decision record

    Each yes or no a person gives is kept, with its reasons.

  • Named sender (pilot)

    In a pilot, outgoing email carries a note naming the sending Org.

The security page covers in full where an Org runs and what it keeps.

Read the security page

Common questions. Three answers cover time saved, new material and trained and managed Orgs.

  • Time saved

    No time saving is measured yet; the delivered record is the evidence.

  • New material

    For now, an engineer adds new material to the Org with you.

  • Trained and managed Orgs

    The Managed Orgs page lists each built Org with its status.

A doubt not covered here can go to the engineer who follows up a request.

Your first Org. Describe the work to cover, and an engineer will plan an Org for it that starts in Nurture.