Why AI seats need a company brain

Almost every company has AI seats by now, and the people using them are faster at some things than they were a year ago. In McKinsey's 2026 survey, nearly nine in ten respondents reported regular use of AI in at least one business function, and eight in ten said it had improved their own productivity.

The company, in most cases, has changed far less. The same survey found that the share of respondents attributing any EBIT impact to AI stayed flat at 37 percent. In the United States, figures published by the St. Louis Fed show 39.2 percent of employed adults using generative AI for work in a given week by the second quarter of 2026, and the time it saves at 2.2 percent of all work hours.

We think the gap has a simple cause. The seats know the internet. They do not know the company.

Why the seats give drafts

Ask an assistant for an offer and it writes a good offer for a company that does not exist. It does not know your prices, your terms, the client's history, the way your company structures a proposal or the discount nobody may give. So it drafts, and a person copies, pastes and corrects. The quality depends on who typed the prompt and how much they knew.

Anthropic's economic research found the same pattern in how businesses use its models: deploying AI for complex tasks may be constrained more by access to information than by what the model can do. MIT's NANDA initiative put it more bluntly last year. According to the lead author of its report on generative AI in business, generic tools stall in companies because they do not learn from or adapt to the way the work is done.

Even the providers say the work is unfinished. A study of ChatGPT Enterprise by OpenAI researchers, updated in August, concludes that firms are still learning how to integrate AI into their workflows, and that the long run value depends on whether scattered individual use grows into a capability of the organisation.

What a company brain holds

A company brain is that missing layer, and it holds four things. Knowledge: documents, emails, offers, contracts, products, prices, past projects and the company's voice, kept current, with the source on every answer. Rules: what each role may see and do, what waits for approval and what may run on its own. Systems: connections to bexio, Microsoft 365, the CRM, the calendar and the company's own tools, so the brain can look things up and write things back. And skills: the way each department does its recurring work, written down so an assistant can follow it.

This is what Twin is. It lives in the assistant your people already use, in Claude, ChatGPT or Codex, or on a private model when the data must not leave: Eigna, built by our partner Eigna Labs, trained on the company's own work and running in its own cloud or on its own servers. Twin is delivered as source code the company owns, so the brain stays with the company when the models change.

Department by department

In sales, the offer from yesterday's call arrives drafted in the company's structure, and the follow up is written in its voice. A person approves both. In support, an answer goes out with the passage from the manual it relies on, a return is handled by the written policy, and anything the brain is unsure of is handed to a person.

In finance, a partial invoice is prepared in bexio through Kontor, open items are followed up on time, and the month end arrives prepared for review. In marketing, next week's posts come on brand, and the campaign report ends with three next steps. The owner gets a Monday briefing on sales, cash and open items, and a decision memo with the source for every number.

New people gain the most. In a study of 5'179 customer support agents, Erik Brynjolfsson, Danielle Li and Lindsey Raymond found that an AI assistant raised issues resolved per hour by 14 percent on average and by 34 percent for novice and low skilled workers. The authors suggest it worked by spreading the knowledge of the most able workers to newer ones. A company brain does that on purpose: the best colleague's knowledge, on every desk, from the first day.

Quality, speed, cost and workload

Quality is the first change: one standard on every desk, with the source on every answer, instead of a result that depends on who asked. Speed follows, because work that took hours starts from a finished draft. Cost falls with less routine, less rework and fewer tools doing the same job. And the workload shifts. People stop moving text between windows and get back to the work only they can do.

That last point matters to us. A brain takes the copying, the looking up and the first draft. The judgement, the conversation with the client, the review and the responsibility for what leaves the company stay with people, and that is where their time should go.

We do not put a number on these gains before we know a company. They depend on how much of its work is repeated, written and checked, which is exactly what a discovery finds out.

Better processes before automation

The research agrees on where the value sits. BCG, drawing on its own client work, splits it like this: about 10 percent of the value of an AI transformation comes from the application itself, 20 percent from data and technology, and 70 percent from workflow redesign, culture, governance and the way people and AI work together. In McKinsey's survey, nearly three quarters of the small group of high performers had fundamentally redesigned workflows because of AI, against one quarter of everyone else. Deloitte's 2026 report found 37 percent of organisations using AI at a surface level, with little change to how the work is done.

So we start with discovery, not with software: the people, the tools, the data and the work still done by hand. Then we sharpen the process before we teach it to a machine. An offer template nobody agrees on, an approval that waits a week, a report built twice in two departments. Fixing those is often the biggest gain, and it makes the automation that follows simpler and safer.

Where to start

A company can begin this week without a new tool. List the ten documents your company writes most often. Write down the rules that live only in someone's head: prices, discounts, tone, who signs what. Note which systems the work runs through and where people copy from one into another. That list is the outline of a company brain.

Then pick one department and one routine, connect the knowledge and the system behind it, and keep a person at the point where anything leaves the company. When the drafts stop needing corrections, move to the next routine. The seats you already pay for will start to work for the company, not only for the person at the desk.