Agents forget. The company should not.

Models reason. They do not know your exceptions. Shared memory is what makes an agent a colleague instead of a tourist.

You can brief a person in a hallway. You cannot brief an agent that way. Y Combinator’s line is the shortest version of the problem: know-how is scattered across heads, Slack, tickets, and databases, and agents cannot operate like that.[[ycLinkedin]]

Amnesia is the default

A new chat is a new hire with no onboarding. Tessera’s write-up of the company-brain wave says it without decoration: agents are born amnesiacs. You paste context. You commit a markdown file. You yell at the model about a customer it should already know. The failure is structural.3

Slite separates the human workaround from the agent constraint. People ask the person next to them. An agent only knows what you hand it, and it forgets between sessions unless something remembers on its behalf.1

Four kinds of memory, not one prompt

The CoALA framework from Princeton (2023) splits agent memory the way cognitive science does: working memory in the current context, episodic traces of what happened, semantic facts about the world, and procedural skill — how to do the job. Most “memory” products keep a chat log and a vector index. That is two layers, loosely.4

A company brain has to hold the other two at organizational scope: what we decided, and how we do this here. Vectorize calls that a shared, enforceable, evolving agreement — not a pile of stored files.2

The pipe is not the brain

Anthropic’s Model Context Protocol (2024) gave agents a standard way to ask an external system for context. That is necessary plumbing. It is not memory. An MCP server in front of a stale wiki still serves stale wiki.5

Foundation Capital’s point still stands: if you are not in the path when the decision is made, you cannot capture the why. Warehouses get the row after the fact. Agents need the precedent.6

Morrow keeps one brain for the team and the agents. Mor proposes the comment, the ticket, the morning report. A person confirms before anything lands back in Slack or Jira. The agent is not a tourist, and it is not unsupervised.

Sources

  1. 1.
    What is a Company Brain? A full guide for 2026

    Christophe Pasquier. Slite, 2026.

  2. 2.
  3. 3.
  4. 4.
    Cognitive Architectures for Language Agents

    Theodore Sumers, Shunyu Yao, Karthik Narasimhan, Thomas L. Griffiths. Princeton / arXiv, 2023.

  5. 5.
  6. 6.
    AI’s trillion-dollar opportunity: Context graphs

    Ashu Garg and Jaya Gupta. Foundation Capital / B2BaCEO, 2025.

  7. 7.

Give your team and agents a brain.

Download free for Mac or Windows — or talk to us about Business and Enterprise.