Y Combinator's "company brain": what YC asked for and what it means
In its Summer 2026 Request for Startups, Y Combinator asked founders to build a "company brain": a system that gathers a company's scattered knowledge, structures it, keeps it current and turns it into something AI agents can act on. The request came from YC partner Tom Blomfield, and the argument behind it was simple — the main thing stopping AI from automating work inside companies is not the models, it is domain knowledge.
The idea spread fast. US searches for "company brain" went from about 10 a month at the start of 2026 to around 590 a month by the summer, and dozens of startups now describe themselves with the term. This article explains what YC asked for, why it resonated, who is building it, and what it means if you run operations rather than a startup.
What YC's Request for Startups says about the company brain
YC publishes a Request for Startups (RFS) each batch: a list of ideas its partners would like to fund. The Summer 2026 list, published in late April 2026, included a company brain among its headline themes.
The core of the request, as YC framed it:
- Domain knowledge is the bottleneck. AI models are capable enough; what they lack is the specific knowledge of how a given company works.
- That knowledge is scattered. It sits in documents, chat, email, tickets, CRMs and people's heads.
- The company brain layer does not exist yet. The tools to build agents exist; the layer that feeds them reliable company knowledge does not.
- "Every company in the world will need this."
The output YC described is not just a search index. It is knowledge collected, structured, kept up to date and turned into something agents can execute — a skills file for AI, not only an answer engine for people.
Source: Y Combinator, Requests for Startups (Summer 2026).
Why YC thinks every company needs a company brain
Agents without context make expensive mistakes
An AI agent that does not know your pricing rules, approval thresholds or exceptions will produce work that looks right and is wrong. The more autonomy you give it, the more that missing context costs.
Knowledge is fragmented by default
No company has its knowledge in one place. Policies live in SharePoint, prices in the ERP, exceptions in a manager's inbox, and the reason behind a decision in a meeting nobody recorded.
Search was never enough
Enterprise search helps people find documents. Agents need more than documents: current data, permissions, and procedures they can follow step by step.
The company brain becomes infrastructure
If every department runs agents, they all need the same source of truth. A company brain is what keeps a quote, an answer and a follow-up consistent with each other.
Why now: what changed in 2026
The idea of capturing a company's knowledge is decades old — knowledge management, intranets and wikis all tried. Three things made 2026 different.
Agents moved from demos to work
AI agents can now carry out multi-step tasks inside business software: reconcile a bank statement, chase a missing document, prepare a quote. Once an agent acts, the cost of it not knowing your rules becomes real.
Connecting AI to tools became standard
Open standards such as the Model Context Protocol (MCP) made it far easier to plug models into business systems. The plumbing stopped being the hard part; deciding what the model should know became the hard part.
Models became good enough that context is the difference
When every company has access to similar models, the advantage comes from what you feed them. Two companies using the same model get very different results depending on the knowledge behind it.
What a company brain needs to do, according to the request
Reading the RFS alongside the products that followed it, five capabilities come up again and again:
| Capability | What it means in practice |
|---|---|
| Ingest | Pull knowledge from the tools a company already uses, not only uploaded files |
| Structure | Organise it around people, customers, products, processes and decisions |
| Stay current | Update as information changes, instead of freezing a snapshot |
| Respect permissions | Show each person and agent only what they are allowed to see |
| Be executable | Turn knowledge into instructions and skills agents can act on |
The startups building a company brain
YC's request accelerated a market that was already forming. A few examples of the approaches now on the market:
- Startups from YC itself. Hyperspell (YC F25) describes itself as the context and memory layer for AI agents; Almanac (YC S26) launched on Hacker News in August 2026 as "AI that knows your company".
- Open source. Supermemory open-sourced its Company Brain — a Slack teammate that remembers what the team says and can act in tools such as GitHub, Linear and Notion — under the Apache 2.0 licence.
- Knowledge-base vendors repositioning. Wiki and knowledge-base tools such as Slite now describe themselves as company brains.
- Enterprise search and memory products. Platforms built around search, organisational memory or decision tracking, such as Glean or Sentra.
- Done-for-you builders. Companies like ours, which build the company brain and the agents on top of a client's existing systems.
We compared ten of these platforms in best company brain software in 2026.
Three common misreadings of the YC request
"It's enterprise search with a new name"
Search finds documents for people. The request describes knowledge structured and kept current so that agents can execute on it. Finding is the first step, not the product.
"It's a chatbot on your Google Drive"
A chatbot over a shared drive answers from whatever files it can read, including the outdated ones. It does not know which source is authoritative, it does not see live data, and it rarely respects permissions beyond the folder level.
"You need to centralise everything first"
The opposite is usually true. Copying all knowledge into a new repository creates a second version of every fact. The durable approach connects each source where it already lives.
What the request leaves open
An RFS is a prompt for founders, not a specification, and a few hard questions are left for the builders:
- Operational data. Much of a company's knowledge is not in documents or chat at all, but in ERP, payroll, rota and stock systems.
- Frontline staff. In retail, hospitality, logistics and manufacturing, most employees are not at a desk and will never open a browser extension.
- Governance. Who approves what an agent does, and how is it logged? In Europe, AI used in areas such as recruitment falls under the EU AI Act, on top of GDPR.
- Ownership. Someone in the business has to own each source and fix it when the brain says it is missing or wrong.
These are the questions that decide whether a company brain works outside a software startup.
Company brain for startups vs established operations teams
Most of the products that answered YC's request were built for a familiar customer: a software startup whose knowledge lives in Slack, Notion, GitHub and Google Drive. That is a real market, but it is not where most company knowledge sits.
In a retail franchise, a manufacturer or a logistics operator, the knowledge that matters lives somewhere else:
- In operational systems — rotas, stock, ERP, payroll, supplier records.
- In procedures — SOPs written for staff who are on a shop floor, not at a desk.
- In people — store managers, HR coordinators and buyers who carry the exceptions in their heads.
The questions are different too. Not "why did we choose this database?", but "who can cover tomorrow's shift?", "is this invoice within tolerance?", "which supplier can deliver by Thursday?".
What it means if you are not a startup
YC's request is useful outside Silicon Valley for one reason: it names the layer most AI projects skip. Before you buy an agent for HR, finance or customer service, the practical questions are:
- Where does the knowledge this agent needs actually live? Documents, systems or people?
- Can it be connected where it is, or would you have to copy it somewhere new?
- Will the agent inherit your permissions, or will someone rebuild them?
- What happens when the agent does not know? Does it guess, or does it tell you what is missing?
- Is it one brain for every agent, or one silo per tool?
How to evaluate a company brain against YC's criteria
If you are assessing a vendor, or your own build, score it against what the request asks for.
| Criterion | Question to ask | What good looks like |
|---|---|---|
| Ingest | Which of our sources can it read today? | Documents and at least one live system |
| Structure | Does it understand our sites, products, customers? | Answers mention the right site or product without being told |
| Stay current | What happens when a policy or price changes? | It reads the current version from the source |
| Permissions | Where do access rules come from? | Inherited from the source systems |
| Executable | Can agents act on what it knows? | Workflows with human approval on risky steps |
| Gaps | What happens when it doesn't know? | It says so and logs what is missing |
How we build company brains for operations-heavy businesses
At Growy, we were building company brains for multi-site and operations-heavy businesses before the term trended, and our approach is the one the RFS points to: connect, structure, keep current, then act.
We connect the documents you already have — SOPs, manuals, policies — and the systems you run on, from ERP and HR to payroll, stock and Microsoft 365. Nothing is copied into a separate Growy database; the brain reads and writes through authenticated connections, and every agent action is logged in the source system. Agents inherit your role-based permissions, and human approval steps sit wherever an action carries risk.
The results come from operations, not demos. At a 100+ store grocery retail franchise, employee self-service resolution rose from 72% to 87% over 24 months as the brain logged every question it could not answer. At a door and window manufacturer receiving 60+ quote requests a day, a quotation agent built on the company's own pricing knowledge brought turnaround from more than 48 hours to minutes.
One brain, 30+ companies
Context: a group of more than 30 companies across technology, hospitality and other sectors, with HR and recruiting centralised at group level. Each company had its own procedures.
Before: 35+ manual steps per hire, vacancy details scattered across emails and chats, CVs screened one by one against outdated job descriptions. Onboarding added another 56+ manual steps.
What we built: recruiting redesigned into 6 agent workflows with SharePoint as the single source of truth for vacancies and candidate files; onboarding rebuilt into 6 workflows across departments.
Governance: because AI is used in hiring, the workflows were built to align with the EU AI Act and GDPR, using only the group's internal best practices, procedures and historical decisions, with humans making the final calls.
Result: 75%+ of HR and recruiting workflows now run on agentic automation; time-to-hire dropped from months to a fraction of that. See how it works in AI agents for HR.
That is the part of the company brain the RFS points towards but most products have not reached yet: knowledge that does not only answer, but runs the process — inside the rules a regulated business has to follow.
Frequently asked questions
What is the YC company brain?
An idea from Y Combinator's Summer 2026 Request for Startups: a system that collects a company's scattered knowledge, structures it, keeps it current and makes it usable — and executable — by AI agents.
Who at YC asked for a company brain?
YC partner Tom Blomfield wrote the company brain request in the Summer 2026 RFS.
Did YC invent the term "company brain"?
No. The phrase was in use before 2026. YC's request turned it into a recognised startup category, which is why search interest rose sharply from April 2026.
Is a company brain just RAG?
No. Retrieval is one technique inside it. A company brain also covers which sources are connected, permissions, freshness and the actions agents can take. See the comparison.
What is a YC Request for Startups?
A list of ideas Y Combinator's partners would like founders to work on, published regularly. It signals where YC sees opportunity, but companies outside the list are funded too.
Are YC company brain startups built for enterprises?
Most early products target software and knowledge-work teams whose knowledge lives in Slack, Notion and code repositories. Operations-heavy businesses should check how a product handles ERP, HR, stock and frontline staff.
Do you need to be a startup to use a company brain?
No. The need is often larger in established, multi-site businesses, where knowledge is spread across operational systems and frontline teams.
How do I get started?
Pick one question that costs your team time every week and the sources it depends on. Book a demo and we will show you a company brain built on your own case.
