Silo Buster Draft front Host an hour

A facilitated programme for leadership teams

AI arrives one person at a time.

Silo Buster gives the room one picture, in everyone's own words, in a single hour.

Each person tells the assistant they already use where they are, what they need and what they could offer. One read across every answer shows the room where it agrees, where it is pulling apart, who is engaged and who is quiet. Then ten working days test whether the habit holds once the facilitator steps back.

Host a walk-in hour See the ten-day test
The fold Ten silos stand in a row, one per person. In the fold one thread runs across the people who agree, a dashed line joins two people pulling apart, one silo is lit as engaged and one is faded as quiet. TEN PEOPLE, TEN ACCOUNTS ONE READ ACROSS THEM: THE FOLD
  • Agreement
  • Pulling apart
  • Engaged
  • Quiet

Illustration. Each column is one person's own account.

The problem it is for

A large share of every week goes on asking.

Each assistant works for one person, inside one company's walls. Connecting systems does not connect people, and the capability already in people stays hidden.

So coordination still fails where it always did. Nobody has a current picture of where anyone is with anything. It is worst where work crosses teams and partners, and every party has its own systems.

Where are you with that?

When are you free?

Who is doing what?

Who decided this?

What Silo Buster is

Everyone's own words. One read across them.

  1. 1

    Each person answers through their own assistant

    About twenty minutes, with the assistant they already use. Nobody installs anything.

    The three questions

    1. 1What more could you offer, and what would let you?
    2. 2How do you work with your AI today?
    3. 3When did something across teams go well, and who was there?
  2. 2

    Each answer lands at a shared address

    In their own words, under their own name. Nobody edits anybody else's. Each person also keeps three lines fresh every week.

    Three lines, every week

    What I am doing
    One thing I need
    One thing I can offer
  3. 3

    One read lays them side by side

    That read is called the fold. It is immediate, it is the group's own words rather than a consultant's summary, and it can be read again next week to see what moved.

    • Where there is agreement
    • Where people pull apart
    • Who understands what
    • Who is engaged
    • Who is quiet

Built for assistants as well as people

One connector, any vendor, nothing to set up but a name. Your assistant can read my weekly lines, leave a note in my space, or find an hour that suits three people. Across vendors, that has not yet been tested with a working group.

If you have run a design sprint, you know the principle

Everyone contributes, every contribution stands, and the shared picture comes from all of them rather than from the loudest voice. Silo Buster adds three things: the contributions are written, they persist after the session, and any assistant can read them back.

How it runs

Two hours in the room. Ten days to see if it holds.

  1. 1One hour

    The walk-in hour

    The leader who sanctioned it opens the hour: permission, not a purchase. Each person answers through their own assistant and writes their three weekly lines. Before the hour ends, the fold is read aloud, so the room sees its own picture the same day.

  2. 2The next day

    The reading

    The answers are read for outliers, not averages, and nobody is labelled. Three kinds of person tend to show up:

    • the people others name when things went well across teams
    • the people furthest forward with their AI
    • the people who could offer more if conditions let them

    Each person gets a short reflection back.

  3. 3One hour

    The action cycle

    A second hour with its own shape, designed by David Pinto. The rules are said first, and anyone can call a rule broken or end the game. Each person says what is at issue until something that matters to everyone becomes obvious. Then each says what they will do about it, and the room confirms it.

    Each commitment is carried by the person who made it, as a card in their own hand, until it is done or set down.

  4. 4Ten working days

    The ten days

    The programme then runs for ten working days, to find out whether the habit holds once the facilitator steps back.

What leadership gets is a read of its own coordination, and of how far its people have got with AI, in their own words and faster than any survey or workshop write-up. Then a test of whether it lasts.

The ten-day test

A good session is not evidence.

The question that matters is whether the group keeps it going once the facilitator has left. So the second week runs without one.

Week one

Facilitator in the room
MonDay 1
TueDay 2
WedDay 3
ThuDay 4
FriDay 5
Day 1 · Mon

Walk-in hour

Accounts written and the fold read aloud. The group arranges one real meeting, timed against the last one arranged the old way: the first evidence, the same day.

Day 2 · Tue

The reading

Outliers found, and a short reflection back to each person.

Days 3 to 5 · Wed to Fri

Action cycle

One cycle on a real pain point the group names, with a commitment by the end of the hour. The weekly lines refreshed, and the fold read aloud at the first meeting.

Day 5 · Fri

Day-five reading

Published: the counts so far, what broke, and what had to be done by hand.

Week two

Facilitator steps back
MonDay 6
TueDay 7
WedDay 8
ThuDay 9
FriDay 10
Day 6 · Mon

Monday prompt

From the convener. Nothing else.

Days 6 to 10

The group carries it

Commitments followed through, or not, by the group. The convener reads the fold at the second meeting. Any contact with the facilitator beyond a bug report is logged, because it counts against the result.

Day 10 · Fri

Result published

Including a loss, if it is a loss.

What is counted

Four counts. The last one decides it.

1Count

Asking where people are

Messages spent asking where someone is with a piece of work, against the previous fortnight.

Pass
Lower
Kill
No change or higher
2Count

Finding a time to meet

Messages and hours from first ask to a confirmed meeting, against the last meeting arranged the old way.

Pass
Lower
Kill
No change or higher
3Count

The cold read

Five plain questions about the group, answered by a fresh assistant from the shared space alone, with no one briefing it. The same five on day one and day ten.

  • Who is free on Friday afternoon?
  • What did the group decide last?
  • Who is waiting on whom?
  • Who offered what?
  • What is blocked?
Scored out of five: kill at two or fewer, pass at four or more. 012345 KILL PASS
4Count
The one that decides it

Still writing in week two

The share of people who write their own weekly lines in days six to ten, with no chase beyond the Monday prompt.

Counts 1 to 3 can all be true in week one with a facilitator in the room. Only count 4 says the habit holds without one.

Count 4, as a share of participants: kill under 20 percent, pass at 40 percent or more. The author's written prior is 20 to 40 percent. PRIOR 20–40% KILL PASS 0%20%40%100%

The pilot's author, James Howard, wrote his prior down: count 4 lands between 20% and 40%, unless the convener reads the fold aloud at the week-two meeting. That is the lever he expects to decide it.

Predictions for every count are written down and dated before day one, so the result cannot be reframed afterwards.

Day ten

Go, reshape or stop. Published either way.

Go

Counts 1 to 3 pass, count 4 reaches 40% or more, and the convener names a number.

Price the next cycle and run it again, wider.

Reshape

The counts pass with the facilitator in the room, but count 4 fails.

The method works and the platform has not yet shown it adds anything a shared page would not. Keep the method; leave the platform as an open question.

Stop

Count 4 under 20%, or any kill switch.

Nobody kept going without help.

Kill switches, on any day: someone writes identifiable information about a client or third party; the convener needs David Pinto in week two for anything but a bug; fewer than half the group writes in any week; the host's information governance lead objects; any Online Safety Act incident.

Why not Teams, Notion or Copilot

The shared space is the notebook. The method is the product.

Inside one organisation, a shared channel or page with an assistant on top can hold much of what people write, and it will be private by default and searchable. Silo Buster does not claim to beat that as a notebook. What leadership takes on is the method: the walk-in hour and its immediate read, the reading for outliers, the action cycle, and a ten-day test of whether it holds.

The shared space starts to matter where work runs beyond one system: across teams on different tools, out to partners, and between people whose assistants come from different vendors.

Three organisations, each with an assistant inside its own walls, joined by dashed lines to one shared address between them. ORGANISATION A ORGANISATION B ORGANISATION C ONE SHARED ADDRESS
Every AI lab is building memory inside its own walls. When your assistant and mine need to work together and we do not work for the same company, there is nowhere for them to meet.

Who it is for

Deciders first. Then their teams. Then their partners.

It starts with the leadership of one organisation: the people who decide, who carry the cost of poor coordination, and who can sanction trying something new. They are also the buyers. A leadership team tries it on itself, gives feedback, and decides whether it is worth more.

If it holds, it widens to the teams they lead, and out to the partners they work with: a multi-agency partnership, a member network or a group of startups with something to coordinate, where the first question is always whose system the work goes in.

The adoption path as three rings: the leadership team at the centre, the teams they lead around it, and their partners on the outside. THEIR PARTNERS THEIR TEAMS LEADERSHIP TEAM START HERE

What is true today, and what is not

Live and open source. Early and unmeasured.

True today

  • The beach, the space Silo Buster runs on, was built by David Pinto over about six months, with AI assistants.
  • It is open source under the MIT licence.
  • It is live, it has survived a full platform migration, and a small community uses it daily.

Not yet

  • Nothing is measured. This pilot is the first.
  • Accounts are public by default, with privacy as a separate setting.
  • Assistants from different vendors working together has not been tested with a working group.
  • Every read and write by an assistant uses paid AI capacity, which is a real running cost.

Before any pilot starts

  1. The group's space is private by default.
  2. The organisation's information governance lead has seen the set-up and not objected.
  3. A one-page data rule everyone agrees: first names and roles only, and nothing about clients, health, pay or anyone's conduct.
  4. Someone other than David has built a working page on it from the public documentation alone, so the test does not depend on its inventor.

Why facilitators carry it

Facilitators are to this what a games master is to a table.

Most members of any online community never write; they read. Nobody contributes to a shared space without a ritual and a visible consequence. That is why Silo Buster starts as a facilitated session rather than a tool you sign up to. If the pilot holds, the method can be handed to facilitators who already run leadership, sprint and team work.

What we are asking

One group of about ten. Ten working days.

A leadership team, or a partnership, with something real to coordinate in the next fortnight and one convener who already runs its meetings.

The first evidence arrives on day one. A reading is published at day five and the result at day ten, including a loss if it is a loss.

James Howard, MLC Innovations Ltd, Sheffield, with David Pinto, who built the beach and designed the action cycle. The ten-day design builds on Matthew Bushell's Capability Gap pilot.

Which is you?

Notes go to the Silo Buster room on the beach, where James reads them. They are public and signed with the name you give.

For your AI

Hand it to your assistant.

Your assistant reads faster than you do. Give it the whole programme as one document, then ask it whether this would work where you are.

Three papers, about 5,300 words. Every line carries its address on the beach, so your assistant can point to the exact one.

Assistant with the beach connector? Paste this

Read proposal:silo-buster on beach.happyseaurchin.com, then the article and the pilot design it points to. Tell me whether a walk-in hour would work for my team, what you would change, and what you would measure.

Or read the source

On the beach each paper is a tree. An assistant can read the outline first and open only the part it needs: experiment 7.4 is count 4.

The papers

Read it in full.

This page shows a snapshot of the beach. Edit a paper on the beach and the hosted page reads the change on its next load.

Paper

Paper