How it works
One brief every morning.
One boundary per client.
Two questions come up more than any others, usually from people who already run several clients at once. What does this thing actually produce? And does it keep my clients apart? Here are both answers, plainly, and then the method that sits underneath them.
01 The artefact
The morning brief
Most AI tools hand you a blank chat window and leave the thinking where it already was, with you. An operations office produces something instead.
One brief, every morning, assembled from your own business context: what needs a decision today, what is sitting with someone else, and what has quietly slipped while you were busy. It is short on purpose. If nothing needs you, it says so.
Morning brief
07:00
Needs you today
- Client A board pack is due Thursday. The revenue reconciliation has been open since last Tuesday.
- Client C renewal quote expires Friday. No decision recorded yet.
Waiting on other people
- Client B finance lead, 3 days, the VAT treatment question.
- Prospective client, 6 days, no reply since the call on the 2nd.
Quietly slipped
- Monthly note not drafted. The last one went out 38 days ago.
- Two invoices past 30 days, neither chased.
Nothing else needs you today.
02 The boundary
Where your clients’ information lives
If you hold several clients, the first honest question about a system that remembers everything is whether it remembers them separately.
It does. Each client keeps its own context. The part that compounds across your whole practice is your own material: how you work, what you have written, how you price, what you have learned. That is the asset. The client material underneath it stays where it belongs.
Shared, and yours
Your methods, templates, positioning, pricing and history. This is the layer that should get better every month.
Client A
- Numbers
- Documents
- Decisions
- History
Client B
- Numbers
- Documents
- Decisions
- History
Client C
- Numbers
- Documents
- Decisions
- History
Nothing moves sideways along that bottom row.
- One context per client, not one pool
- Each engagement keeps its own material. Work on one client does not draw on another client's numbers, documents or history.
- Nothing leaves without you
- Drafting, summarising and preparing happen freely. Sending, publishing, paying and deleting wait for you. That boundary is set up on day one, not added later.
- A local option, if you want it
- Where the material is sensitive enough, the whole thing can run on a model hosted on hardware you control, so nothing goes to a third party at all.
03 The method
Provenance, permissions, pulse, playbooks
Underneath both of those answers sits the same four-layer method. It is built in this order on purpose, because each layer is close to worthless without the one before it.
Most AI projects start at the fourth layer, with the workflows, because that is the part people can picture. Then they fail quietly, and everyone concludes the tools are overrated. The tools were fine. The first three layers were missing.
- 01
Provenance
Where the truth livesBusiness memory that is source-backed. Every answer can be traced to the document, decision or conversation it came from, so you can check it rather than trust it blindly. Memory without provenance is just a confident guess.
- 02
Permissions
What it may do aloneAn explicit boundary between what an agent does freely and what waits for a person. Reading, drafting and preparing are safe. Sending, publishing, paying and deleting are not. Set on day one, not bolted on after something goes wrong.
- 03
Pulse
When it happensThe daily and weekly rhythm. A morning brief, a chase review, a weekly reset. This is the layer that catches the quiet failures: the thing nobody owned, the reply that never came, the invoice nobody chased.
- 04
Playbooks
The work itselfYour own repeated work, turned into workflows that run the same way every time. Built from what you actually do, not from a template. This is the layer everyone wants first, and the one that fails without the three above it.
We build all four for you, then teach your team to run them. The course is the same four layers, in the same order.
If you want to see the same thing running on a real business rather than an illustration, the case study is the whole of one.