What this is
Company Brain is infrastructure for capturing, organizing, and serving a company's internal knowledge to AI agents. Not a knowledge base tool. Not a documentation platform. A data layer that makes company-specific automation possible.
The core problem: AI agents need to operate inside a company's context. They need to know how your sales process actually works, not how it theoretically works on the website. They need to understand your pricing exceptions, your customer relationships, your internal politics, your workarounds. That knowledge exists. It's just distributed across email, Slack, spreadsheets, people's mental models, and legacy systems. Pulling it together manually for every new agent you build is not scalable. Neither is training models from scratch for each company.
Company Brain automates the ingestion, normalization, and retrieval of that knowledge. A company connects its sources (email, Slack, CRM, databases, documents). The system extracts context, deduplicates, and builds a queryable knowledge layer. When you spin up an AI agent, you give it access to that layer instead of copying instructions into a system prompt. The agent learns the company's actual operating model.
This is not ChatGPT for your Slack. It's not semantic search over your documents. It's a knowledge compilation system that produces structured, actionable context for agents.
Why it's interesting
The economic unlock is real. Companies today spend 200-500 hours per new automation project just gathering context. That's 5-12 person-weeks of setup time before the agent even starts working. Once the knowledge layer exists, that drops to hours. Spinning up a second agent becomes trivial.
The companies that solve this internally will move 10x faster on automation than their competitors. They'll operationalize AI in 2 months instead of 8. Wes's Sales Connector inboxing agent needed 300+ hours of manual knowledge extraction. A working Company Brain instance would have cut that to 20 hours.
The platform also solves a liability problem. When knowledge lives in people's heads and email archives, you can't audit how decisions were made. You can't ensure consistency. An AI agent trained on a formalized knowledge layer is more defensible, more governable, more expensive to replace.
Why a landing page would fail
A landing page sells the dream: "Connect your tools and automate everything." The reality is much messier.
Companies will ask: what does "knowledge" actually mean for my business? The answer depends on their operations, their AI maturity, their data hygiene. There's no one-size-fit-all answer. A startup with 10 years of email chaos and a compliance-heavy enterprise with structured data have completely different problems.
The second issue is trust and control. Pulling knowledge out of email and Slack feels invasive. Companies will demand guarantees on what's stored, who can access it, how long it's kept. Those conversations can't happen on a landing page. They need a sales cycle.
The third is data quality. A company's internal knowledge is often contradictory, outdated, or embarrassingly unstructured. Clients will discover that their "knowledge" is a mess of conflicting procedures, stale emails, and tribal memory. They'll blame the product before they blame themselves. You need a real relationship to get through that.
The realistic shape
Company Brain is a B2B SaaS product with a technical sales motion.
Architecture: A backend that handles ingestion (connectors for Gmail, Slack, PostgreSQL, Salesforce, custom APIs), normalization (deduplication, entity extraction, relationship mapping), and retrieval (semantic search plus structured queries). An admin dashboard for managing sources, reviewing what got extracted, setting permissions. An API for agents to query the knowledge layer. Probably a backend service that handles long-running indexing jobs. Two to three person engineering team.
Sales: Start with 5-10 design partners. Companies that are already building AI agents internally and are desperate for a better way to feed them context. Rui Pimentel at Sales Connector, Ally's network, companies already running multiple agents. Charge $2-5K per month with volume tiers. Target 20-40 customers in year one.
Team: Founder (product, vision), one senior engineer (backend, data pipeline), one engineer (frontend, integrations). A part-time advisor with domain knowledge (someone who's built knowledge systems at scale). You can get to 5 customers with three people.
Capital: $500K should be enough to get to product-market fit. $50K as initial engineering budget, $150K for founder salary, $100K for go-to-market, $100K reserve. You're not raising a Series A until you have 10+ customers with 80%+ net retention.
Six-month milestones: Build ingestion for Gmail, Slack, PostgreSQL. Implement basic normalization. Get first design partner live. Second design partner. Third design partner. Measure retention and CAC. Decide whether to continue or pivot.
Honest 12-month case
Best case: You get 8 customers paying $3K/month each. Revenue run rate is $288K. Growth is strong because automation is genuinely urgent and your product actually works. You've found product-market fit. You raise a Series A on the strength of retention metrics and expand the team.
Base case: You get 4 customers, $600/year. The problem is real but the addressable market is smaller than you thought. Many companies decide to build this themselves. You're still viable but growth is slow. You need to add more connectors and figure out what wedge actually drives adoption.
Worst case: Companies want this to be easier than it is. They expect it to work like Zapier. You build deep connectors instead of shallow ones. That becomes 12 months of engineering with no revenue. You run out of runway. Product dies.
Kill criteria: No design partners by month four. Churn over 10% per month. Your own agents (testing the product internally) don't improve retention by 3x. Any evidence that customers are building internal solutions instead.
Five questions to answer before committing
1. Can you get three design partners to commit 20 hours each on setup in month two? This tells you if the problem is real enough that people will actually use the product.
2. Can you build connectors for the top five sources faster than your customers can write one-off scripts? If the connectors are slow to build or worse than custom code, the product has no leverage.
3. What does "success" look like for your first customer? Is it "my agent works faster" or "my team stopped being the bottleneck"? If you can't define this clearly before you start, you'll fight about it in month five.
4. Who owns knowledge quality on the customer side? If the answer is "the product," you lose. If the answer is "the customer," can you build enough tooling to make that realistic?
5. Is the real moat the ingestion layer or the retrieval layer? Building connectors is hard. Building good retrieval is also hard. You probably can't do both well at first. Which one actually creates defensibility?