McKinsey's new article opens on a Tuesday at a global insurer. At eight in the morning the CEO asks the company's AI operating system how exposed she is to a proposed cap on margins. By four that afternoon, analyses from seven countries have rolled up into her view, with the assumptions behind every number visible.
Read it as the founder of a sixty-person company and you will recognise the ambition and very little of the cast. You have no country heads and no actuaries. You have a co-founder, a part-time CFO and a chair who sees the numbers six times a year. So where would an AI operating system land first in a company like yours? Our answer is the board.
The company in this piece is illustrative. No client's data appears in it.
What McKinsey means by an AI operating system
On 13 October McKinsey published The AI operating system: A new way for CEOs to run the enterprise. It describes a layer that runs continuously, holds the company's context, reasons across functions, recommends or executes decisions and learns from outcomes. The starting point is a gap in McKinsey's own 2026 State of AI survey. Forty percent of respondents at large companies report scaling AI agents, and only six percent attribute at least five percent of EBIT to AI.
Most of the article is about how management changes once information stops being scarce. Leaders decide which decisions agents may make or carry out and which need human judgment. They manage by intervention instead of by calendar. And they keep the company's knowledge portable, under one rule: "Own the context; rent or replace the technology."
Why the board, in a company your size
McKinsey writes for enterprises with layers to remove. A founder-led company has few layers, and its important decisions already pass through one place.
The board is where decision rights are written down before any AI arrives. The articles, the shareholders' agreement and the list of reserved matters say what needs the board and who signs. An AI operating system needs rules like these to work inside, and at the board they already exist.
The board record is also the context that compounds. Each decision, the reasons for it and what happened next is the institutional knowledge McKinsey tells companies to protect. In a sixty-person company it lives in the founder's inbox and a folder of old packs, until a buyer or a new investor asks for it.
And the board is where a wrong answer costs most. A figure in a pack is a claim the directors will be held to. If an AI system is going to touch the company's decisions, the place to give it rules first is the place that already has the strictest ones.
What that looks like in Boardverse today
Boardverse is the AI operating system for the board. Five of the things McKinsey describes are live in it now.
Agents work inside decision rights
McKinsey asks leaders to decide what agents may do on their own. In Boardverse that is settled by the type of record, before an agent is connected. An agent acting for a Founder can update the competitor list, the product backlog or the sales pipeline, and the change is filed at once. A decision, a risk, a resolution or a finance pack is different. The agent's submission waits as a proposal until a different named person confirms it, never the one who proposed it. For an agent's submission that person is a Founder other than the one the agent acts for. A board resolution needs the Chair and a Founder, and if the text changes between the two signatures, the first is invalidated.
You own the context and rent the model
Ask Poseidon, the assistant inside Boardverse, runs on the model you choose, Claude, OpenAI, Azure OpenAI or Gemini, with your organisation's own key. Boardverse trains no model on your record. Agents connect over MCP, an open standard, so the assistant your team already uses reads the same record and files to it under the same rules. That is Claude, ChatGPT, Gemini CLI or Google Antigravity today, and others through MCP. If you change model next year, the record stays where it is and the new model reads it. Why the record matters more than the model makes the longer case.
The record learns from outcomes
McKinsey's system improves as it captures decisions and what followed. Boardverse does this with Decision Memory, released on 21 October. Before a decision is proposed or decided, the board sees earlier decisions on the same question and how they turned out. On the check date the decision's owner answers one question, did it work, and overdue checks show on the Dashboard and in the board book. No model is retrained. What improves is the record the next decision is made against.
Answers cite the record
The insurer's CEO could see the assumptions behind every number. Ask Poseidon answers from confirmed records only and shows the module and record each answer came from. It reads only the modules the person asking is allowed to open.
It keeps running between meetings
Boardverse runs governance sweeps between meetings, so a director's question on a Tuesday is answered from a current record, not one brought up to date at pack time. Agent routines log every run, including the runs where nothing changed, so you can see what was checked as well as what was filed.
Where to start this quarter
- Write down the decisions reserved for the board and who confirms each. Your articles and shareholders' agreement are the first draft.
- Agree in writing which records an assistant may change on its own and which wait for a person. In Boardverse the split is already set: operating lists are filed at once, governance records wait.
- Put each material decision on record with a date to check the outcome, so the next decision has something to learn from.
- Connect the assistant your team already uses. Ask it ten questions whose answers you know before you trust it with one you don't. How to ask your own AI for a number you can trace sets out the habits.
McKinsey closes by telling CEOs to redesign decision rights and management rhythms, and warns that too much caution can stall the work. In a founder-led company the board is the smallest place to do both, and it is the record a buyer will read.
Check out Boardverse, or start with the five-answer diagnostic.