What is a company brain?

In most small and midsize companies, knowledge is spread across records, people's heads and message threads, and the company runs because people remember who knows what. That is Blomfield's description too: some of it lives in people's heads, some in old email accounts, chat threads, support tickets and databases.

In an HVAC service company it looks like this:

When the senior technician is on leave or the sales manager moves on, part of that knowledge goes with them. For an AI agent it is starker still: the agent cannot ask anyone "what did we do last time?"

Document search answers "what has been written about this?"; a company brain answers "what do we do in this situation?" Blomfield describes it as a living map of how a company works: how refunds get handled, how pricing exceptions are decided, how engineers respond to incidents.

When an agent does a piece of work, it looks at two things: what the record says, and what the company does in that situation. Blomfield calls this an executable skills file that AI systems use to do the work safely and consistently.

A service request runs like this:

• A customer reports a fault. The agent checks the equipment record: still under warranty, last serviced four months ago.
• The company's procedure (example): no labour charge on units under warranty, parts come out of the stock in the service van.
• The agent checks the calendar of the technician who did the last service and suggests the assignment.
• Depending on the company's approval rule, the suggestion is saved directly or goes to someone for approval.

When the procedure is on record, the agent takes the same step every time the same situation comes up. When it is not, the agent gives a general answer or asks you.

A company brain starts with getting knowledge on record; choosing the tool comes second. Five steps:

• Move records into one structure. Link customers, equipment, quotes, work orders and invoices to the same customer record instead of spreadsheets, WhatsApp and separate programs.
• Write down recurring decisions. Who gives special pricing, which amount needs approval, what gets charged on a unit under warranty?
• Turn procedures into steps. Define flows such as "when a quote is accepted, open an order and notify the warehouse" as trigger, condition and step.
• Set the approval line. Decide what the AI saves directly and what waits for approval.
• Close the loop. Review the agent's suggestions, fix what went wrong and update the procedure. In her piece on the same list, YC partner Diana Hu describes this as moving a company from an open loop to a closed loop: the system watches what is happening, compares it with what should be happening and adjusts.

In PureField, the recorded part of company knowledge sits on one platform. CRM, field service, ERP, purchasing, finance and HR use the same customer, equipment and product records, and quotes, work orders, service reports and invoices are linked to them.

• Copilot answers questions you ask in everyday language from these records. When a question needs your records, it opens a management meeting with the sales, finance or service personas involved.
• [Workflows](/en/workflow-automation) define your company's procedure step by step: they start when a record is created or at a set time, branch on conditions, create or update records, and send emails and panel notifications.
• Digital employees watch work such as collections, quote follow-up and service planning; by default they prepare a suggestion and wait for your approval.
• Approval rules depend on the kind of change: deletes, missing fields and content read from outside sources go to approval, and purchase orders are always approved by someone on your team.

Copilot and digital employees run on PAIOS (PureField AI Operating System). See which agent works in which module on the AI agents page.