Options for building a company brain

Give agents a working understanding of the business

A company brain is a shared, maintained understanding of how your company works, available to the people and agents doing the work.

It brings together the background that normally lives across documents, messages, meetings and people's heads: who's involved, how projects connect, what customers need, what's been decided and why.

Think about a colleague who has been properly onboarded. You still give them a brief. You still answer questions. But you don't have to explain the whole company before they can make a useful contribution. They know where to look, recognise important constraints and remember what happened last time.

That is the standard we should be working towards with agents.

They should start with a useful picture of the business, investigate the details they need and leave useful work behind for the next person or agent. Over time, the company should become better at using AI because its shared understanding improves.

The bigger opportunity is handing over more of the work

For a quick question, good search may be enough. If you want to find the holiday policy or check a date, finding the right document can solve the problem.

The demands change when you ask an agent to work through a customer problem, develop a launch plan or investigate a product opportunity over a longer session.

It has to make a series of connected judgements. Which customers matter here? What have we already tried? Which commitments need protecting? Who needs to be involved? An agent can find relevant documents and still miss how the business fits together. If it starts from an outdated decision, that mistake can shape everything it does next.

Multiply that across a team. One agent builds a launch plan around your old target customer while another writes messaging for the new one. Each piece of work might look convincing on its own, but they are pulling in different directions. As more agents work in the background, those differences become harder to spot.

A company brain gives agents across people, tools and sessions a shared starting point for those decisions. They can spend more of the session moving the work forward, with less time rebuilding context or waiting for you to supply it.

We believe agents will take on more responsibility and work for longer with less supervision. Giving them a dependable understanding of the business is part of making that possible. The better question to ask of a company brain is how much useful work it helps an agent carry through.

Lovable is already showing the demand

Bruno Petersen recently shared that Lovable had built its own company brain. He reported that more than 60% of the company used it every week and that it had served over 200,000 requests, including requests from agents. Bruno's post.

For me, that is a strong signal. A company building at Lovable's pace chose to invest in shared company knowledge, and people are using it regularly.

You do not need to build exactly what Lovable built to benefit. There are a few ways to give your own team a company brain.

Build one yourself

Building a company brain means creating a shared understanding of your business that agents can use and that stays useful as the business changes. There are a few connected problems to solve:

• Choose and connect your sources. That might mean Slack for ongoing conversations, Notion for documentation, Linear for tickets and GitHub for code. Decide which information matters and how to bring it in as work happens.
• Build a coherent picture. Work out how people, customers, projects and decisions relate across those tools. A customer conversation, a ticket and a code change may all describe different parts of the same piece of work.
• Handle conflicting information. What happens when a document says one thing and a conversation says another? Decide what to trust, distinguish suggestions from agreed decisions and flag uncertainty when someone needs to resolve it.
• Keep the understanding up to date. When a source changes or is deleted, identify which explanations and related information need updating. A changed customer commitment may affect the account history, a project plan and the priorities an agent works from.
• Give agents a way to navigate it. They need a useful starting point and a way to find the detail for their task. Keep explanations linked to their evidence so agents can check the original material when the prepared understanding is not enough.
• Respect access permissions. Decide what each person and agent can see, including in explanations drawn from restricted sources. Work out what happens when access changes.
• Carry useful learning forward. Give people and agents a way to save findings, decisions and corrections for future work, while keeping tentative ideas separate from accepted company knowledge.

Someone also needs to own the ongoing operation: spotting failed connections, managing costs and checking whether agents actually do better work with it. Building internally can make sense when you need that control and have the team to maintain it. Lovable is a good example of a company choosing that route.

Start with an open-source option such as GBrain

GBrain gives you a substantial head start. It already includes ways to organise knowledge, connect sources, preserve useful information from conversations and maintain understanding over time. It also supports company use.

That makes it a real option for a technical team. You can adapt it to your own way of working and control how it runs.

You still need someone to set it up, connect the sources you care about, manage access and keep it running. Some connections have ready-made instructions; others need work to fit your company. Its Slack setup guide, for example, walks through creating workflows for reading selected channels and filing useful information.

The trade-off is straightforward: you get flexibility and a lot of existing functionality, while someone on your team owns the resulting system. For some companies, that is exactly what they want.

Choose a product that does the ongoing work for you

The other route is to use a company-brain product, so your team can focus on what agents do with the knowledge.

Look at how it fits the way your team works: the sources it connects to, the agents that can use it and how it handles new information, corrections and work carried over from previous sessions.

Try it on a task that needs both an understanding of the business and some investigation. Can your agent make a useful start, then follow the evidence back to the original material when it needs more detail? That combination becomes particularly valuable during longer pieces of work.

This is a central part of how we think about Pensieve.

Pensieve gives your agents a map of the business, with the detail behind it

We built Pensieve for teams that want a company brain without making its construction and upkeep another internal project.

You connect selected information from the tools where your company works. Pensieve organises it into a shared map of the business: people, customers, projects, processes and decisions, with useful explanations of how they fit together. It maintains that understanding as connected information changes. How the context layer works.

An agent preparing your launch plan can start with the product's purpose, the customers it matters to, the decisions already made and the work it depends on. It has a place to begin and a way to find the right detail.

This is the value of preparing some understanding in advance. Every agent should not have to repeat the same investigation just to discover how your business fits together. The aim is to spend fewer tokens reconstructing the background and less of your time explaining it.

The map also keeps a route back to the evidence. Your agent can read the original material behind an explanation or search the underlying sources directly when it needs to go deeper. It can use its other tools to check live figures or take the next step in the work.

That combination matters: a useful starting point, with room to investigate beyond it.

Your team can inspect and edit the shared understanding too. If something is wrong, the correction becomes part of the maintained map that connected agents can consult. You are improving the starting point for future work across the team, instead of fixing the same misunderstanding separately in each conversation.

Conversations are a valuable source of company context in their own right. They are where you investigate problems, test approaches, explain why something matters and work out what to do next. Much of that understanding never makes it into a document.

You can choose to add your Claude Code and Codex conversations to your company brain, making that work available for agents to analyse and build on across sessions and across your team. The next agent can find what was already investigated, which approaches were tried and why you changed direction, without asking you to reconstruct it all.

This is how learning can compound into something closer to what we would call experience. Your agents have a growing record of how your company has worked through real problems, alongside its documents and current priorities. Each piece of work can leave the next agent better informed, and the understanding your team builds with AI becomes an asset the company keeps.

Pensieve brings that maintained map, source access and shared context into one product, available to the agents your team already uses. We take on the work of building and operating the system around it. Your team decides what to share and puts it to work.

Choose based on the work you want to hand over

Take a real task that normally needs several rounds of explanation. Try it with the company brain you are considering.

Does the agent understand enough to make a useful start? Can it find the detail behind an important decision? Do you spend less time correcting its picture of the business? Can another session pick up the useful work afterwards?

Those are the outcomes that matter.

If you want to build and run your own system, there are good starting points. If you want your team to have a maintained company brain and get on with using it, that is what we are building Pensieve for.