Member News

When a company thinks about AI, it usually starts with the use case: an assistant for the sales team, a tool that drafts bids, a chatbot for customers. The use case is the part people can see, so it is where the attention goes. I think it is the wrong place to start. 

A use case is only as good as the data it draws on. In most organisations that data sits in several systems, in different formats and at different levels of quality. Gartner predicted that, through 2026, organisations would abandon 60% of AI projects that lack AI-ready data. Starting from the use case turns each project into intensive, custom work.

I came to AI after years leading transformation at banks including Goldman Sachs and Barclays. At Codefully, across many AI projects, I see the same pattern repeat: systems work in silos, and the dots between them stay unconnected.

That is why the brain should come first.

What an AI brain is 

An AI brain is a live layer that connects a company's documents, systems and teams, with access rules that control who sees what.

The easiest way to picture it is as your company's library. A company keeps documents in Google Drive, SharePoint and other folders, along with calendars, meetings and long email threads. The brain pulls that material into one place on top of the company. You can ask it what the team agreed in a meeting a month ago, or what a document says. 

I define the brain by what it connects. It joins data and intelligence in one layer, links to your systems, and lives and breathes as your teams work and progress.

The brain follows role-based access. A company has communications, marketing, development and sales teams, and each team needs different information. A developer has no need to see a new sales prospect until the deal becomes real. The company website is open to the team, while a colleague's email stays private. Each person reaches the information their role requires. 

I see an AI brain as the easiest way to put the people and the systems of a company on the same page. 

The brain and its arms and legs 

The brain is the foundation, and the use cases are its arms and legs. The brain holds what the company knows. The use cases put that knowledge to work: answering a technical bid, advising a junior member of staff, automating part of a workflow. With the brain in place, each use case builds on the same base.

The brain works at two levels. At the first, it gives people access to information. Teams keep documents in different places, formats and access points, and the brain adds intelligence on top, so people find the right information at the right time in a fast, organised way.

At the second, it puts that knowledge to work. Because it understands the data and sees across the organisation, it can carry out tasks such as drafting answers or completing part of a workflow. People spend less time gathering information and more time on the work itself. 

Why an AI brain needs memory and retrieval 

For a brain to work at either level, it has to do two things at once: remember what the team has discussed, and find what the company holds. A language model does neither on its own.

A model learns up to a fixed point in training, then stops, so each conversation starts from that frozen snapshot. It can also hold only so much text in view at once. Leading models now offer windows of up to one million tokens, but in our experience answer quality starts to degrade past about 500,000. Chroma tested 18 leading models and found that reliability falls as the input grows. The software around the model therefore has to choose what it sees: the relevant part of a document, the key points of an older conversation.

Most tools today handle one side of this. Some improve long-term memory but cannot retrieve from your documents. Others use retrieval augmented generation (RAG) to search documents, and their memory stays weak. Microsoft Research found that baseline RAG struggles to connect the dots across separate documents. A brain needs both in the same system: memory of last  week's conversation, and retrieval of a contract from last year.

Keeping the knowledge current 

Memory and retrieval only help while the knowledge behind them stays true. A tool that copies a document into its own store falls behind the original. Edit a spreadsheet in Google Workspace, and the copy keeps answering from the old version until someone uploads the new one. A brain reads the live source, so it needs the rights to your Google Workspace or Microsoft tenant. 

Data rot is the second risk. A company collects a large amount of information over time, and without curation the useful data ends up beside outdated material. The first step against data rot is to let people review what the brain holds and correct it. 

That is why I describe the brain as a living layer. It follows teams, people, departments and documents as they change.

The market is moving towards the brain 

The large AI vendors are building the same idea into their products. ChatGPT's company knowledge connects to tools including Slack, SharePoint, Google Drive and GitHub. Claude's enterprise search covers Slack, Microsoft 365 and Google Workspace. Both find documents across those tools, but their memory belongs to each user or project, not to the company as a whole.

The gap comes down to design. These products serve a general audience, and a memory system tuned to one company does not suit another. A company brain is built around one company and tuned to how it works.

Start with the brain 

Small and large organisations both stand to gain from an AI brain. Systems that work in silos leave the dots between them unconnected. An AI brain connects those dots in a short time, and lets people connect them. With data and intelligence at the centre, a team gets more done with less effort. 

Start with the brain, and build your use cases on top of it.

These ideas shaped Brainfully, the AI brain Codefully is building for organisations. Check out brainfully.ai to learn more. 

Author: Neil Sheth, Co-founder and CEO of codefully