New whitepaper · 10 November
How to build a company brain
Every leadership team wants the same thing from AI: an assistant that answers the way the company would, with its decisions, its clients and its reasons. That is a company brain. It has three parts. This paper explains each one, gives away the use cases and one skill written out in full, and says which part to leave to someone else.
In short
A company brain is an AI that answers the way your company would, with its decisions, its clients and its reasons. It has three parts: maintained context, a foundation model, and your own agents and skills. In Parallel’s whitepaper explains each part, gives away the use cases and one skill in full, and says which part not to build yourself.
18-page PDF · free · published Tuesday, 10 November 2026
Part one · The formula
Context + models + skills.
A company brain is not a product you buy or a model you train. It is what you get when three things are put together and kept together. Two of the three you can rent or write yourself. The first has to be maintained, every day.
Context
What your company has decided, promised and learned, kept current and scoped to each reader.
Maintained
A foundation model
Claude, ChatGPT, Copilot or your own platform. It supplies the reasoning, and knows about your firm only what it is given.
Rented
Agents and skills
Your playbooks, written as instructions an AI can run: how you synthesise interviews, hand over a client, prepare a board pack.
Written by you
A company brain
An assistant that answers the way your company would, for every person and every agent allowed to ask.
Why the first line decides the rest
The model and the skills are only as good as the context they are given. The first line is also the only one that changes every day. A model is replaced a few times a year. A skill changes when your playbook does. The context changes with every meeting. So the three parts need three kinds of ownership: one is a purchase, one is a writing job with an owner, one is a maintenance job.
The numbers
The model is rented. What your company knows changes every day, and nobody is maintaining it.
Use AI at work
Adoption is not the gap. Daily AI users report 20.3 coordination hours a week, against 9.1 for those who use none. An association, not a cause.
Brief it by hand
Of managers who use AI, the share who must load context before the assistant is useful. Over half spend three minutes or more on it each session.
Live only in memory
The share of teams whose meeting decisions are written down nowhere. A further third rely on someone’s personal notes.
The fragmentation multiplier
Weekly coordination hours for teams on more than ten tools, compared with teams on one to three.
The shared-context dividend
Hours a week managers expect back if their AI could read the team’s shared context. An expectation, not a measurement.
Per manager, per year
4.4 hours × 46 weeks × $85 an hour. The same on every route, so it settles nothing on its own: the cost side does.
Figures from The Coordination Tax Index 2026, In Parallel’s survey of 247 managers in five countries, June 2026. Hours are self-reported; read them as directional.
Part two · The raw material
Context data is toxic before it is golden.
Your systems of record already hold what happened. What a company brain lacks is why, and what was promised. That is mostly spoken, so the first instinct is to record everything and point an AI at the pile. That instinct produces the toxic form.
Raw: the toxic form
Everything in one file
The pay discussion, the client’s name, the remark about a colleague. Whoever has the file, person or agent, has all of it.
No sense of time
It records what was said, not what was later reversed. An assistant repeats a decision overturned a fortnight ago, and sounds sure of it.
No sense of authority
A passing remark gets the same weight as an agreed commitment, with no word on who had the right to make it.
Personal data, piling up
A recording is personal data about everyone who spoke. The pile grows with every meeting, and so does what you must answer for.
Refined: the golden form
Decisions, with owners and reasons
Along with the actions, risks, open questions and learnings around them, each traceable to the meeting it came from.
Linked over time
Each decision carries what it replaced, so a reader can tell the current answer from the old one.
Sorted by sensitivity
Raw records, observations, storylines and conclusions are shared at different levels. The more distilled, the wider it travels.
Scoped to each reader
A person, or an agent acting for that person, reads only what that person may see. The transcript is deleted on your schedule.
Share the answer, not the archive. Everyone gets the same truth, and nobody gets more of it than they should.
Part three · Use cases
What a company brain is for.
The use cases we see work first. We give them away because the skill is not the hard part: each is a page of instructions, and the paper writes one out in full. Each depends on the same thing underneath.
The status update that writes itself
“What changed in this project since Monday?”
Handover without the handover meeting
“What has been decided here, by whom, and why?”
The decision lookup
“What did we agree on pricing for this client, and is it still current?”
Meeting preparation
“Brief me for tomorrow’s steering group.”
Client and board reporting
“Draft this month’s report for the client.”
Interview synthesis
“What did the twelve customer interviews say about onboarding?”
Commitment and risk review
“Which commitments from last month are still open, and who owns them?”
Agents that act with context
“Draft the proposal, using what we agreed with this client.”
Part five · Build, bundle or buy
Three ways to get there. They fail differently.
Build it
Your own pipeline
Where it fails
The raw pile arrives first. Sensitivity, supersession, deletion and audit come later, and need a team that stays.
Right when
Nothing may leave your own estate, and you can staff it for years.
You own the upkeep.
Use the bundle
Your assistant or notetaker
Where it fails
It mostly reads what is already written down. The memory serves one assistant: the next starts from nothing.
Right when
One vendor, one assistant, and few walls needed between teams or clients.
One vendor owns your context.
Buy a substrate
A maintained layer
Where it fails
A new dependency and a new processor of your most sensitive material. Only as good as its extraction.
Right when
More than one assistant, walls between clients or projects, engineers you would rather put on skills.
You must be able to leave.
You build the brain. We maintain the substrate.
The brain is yours: your AI platform, your models, your agents and your skills. That is where your firm is different. The substrate under it is a maintenance job, and it is the job In Parallel does. You should be able to check that, and to leave — the paper ends with twelve questions to ask before you trust anyone with this, us included.
The full paper
Published Tuesday, 10 November 2026
The full 18-page paper comes out on 10 November. Leave your work email and we send it to you that morning, before it is anywhere else.
Part three
One skill, written out
Use case 1 in full, as we would hand it to an assistant: when it runs, what it reads, what it writes, the rules, and the check before it ships.
Part four
The build list
Six working parts — capture, connect, structure, run skills, govern, share — and why none of them stays done.
Part five
The demo is easy
What a home build gives you on day one, what it still owes you, and how to do the sum on what it costs to keep.
Part six
Who does what
What stays yours — the platform, the models, the skills — and what a substrate should keep running for you.
Part seven
Twelve questions
What to ask before you trust anyone with this, on the record, the walls, the data and the exit. Ask them of us.
Part eight
Limits, and where to start
Five things a substrate will not do, and a thirty-day pilot with one team and two use cases.
Send it to me on 10 November
One email, the morning it is published. Nothing else.
We respect your privacy. Privacy Policy
The webinar series
See it built, one use case a week.
Four one-hour webinars, every Wednesday from 18 November. Each one shows one team using shared context for one kind of work: a 40-minute session on the case and the product, then 15 to 20 minutes for your questions. Free and online.
-
Wednesday, 18 November 2026
How to build a company brain: three parts, and the one that is hard
“Why does our AI still not know what we decided?”
-
Wednesday, 25 November 2026
Projects that do not drift
“What was decided last week, and what could slip?”
-
Wednesday, 2 December 2026
Customer success on shared context
“Which customers are waiting on us right now?”
-
Wednesday, 9 December 2026
Build or buy, and where management starts
“Where do we stand against the plan this week?”
Frequently asked
About the whitepaper
Is it free?
Yes. Enter your work email and the PDF is yours. No follow-up sequence.
When is it published?
On Tuesday, 10 November 2026. Leave your email and we send it to you that morning.
Does it name vendors?
No. Part five describes kinds of product — a home build, the memory bundled with an assistant, a bought substrate — as we understand them in October 2026. We sell the third, so the paper includes the questions to ask us.
Where do the numbers come from?
From The Coordination Tax Index 2026, our own survey of 247 managers in five countries, June 2026. The paper adds no new data. Hours are self-reported, so read them as directional.
Is there a webinar?
Yes. How to build a company brain is a free series of 4 one-hour sessions, every Wednesday from 18 November. Each one takes a use case from the paper and shows it live.