Companies Are Already Stupid AGIs

The real idea here is that a company already works like a slow, messy reasoning machine, and AI matters when it replaces the handoff layer between people. In most firms, information lives across meetings, emails, CRM notes, decks, and chat threads, then gets reassembled by analysts, PMs, associates, and managers into a recommendation or decision. The claim points to software that listens, indexes, remembers what was surprising, and reuses it in the next workflow automatically.

• In the Tribe example, the concrete bottleneck is underwriting. A partner talks to a founder, pulls company metrics, normalizes them against past deals, adds qualitative notes, and turns that into an investment memo. The proposed AI system does the indexing, comparison, recall, and draft synthesis that junior staff and middle managers normally handle.
  1 sacra
• This same pattern shows up in social products. Reddit's upside was framed as taking a huge corpus of niche discussions, understanding what each user wants, and remixing the best threads into a better package. That is the same company as AGI idea applied externally, turning raw distributed knowledge into a tailored output.
  1 sacra 2 sacra
• The broader implication is that AI agents are less like a new app category and more like software employees sitting on top of existing systems. Related research on OpenAI and agent infrastructure describes agents as tools that execute multi step work, while integration layers like Ampersand handle the messy system access, retries, auth, and data mapping needed to make them useful inside companies.
  4 sacra 5 sacra

This is heading toward companies where the default unit of work is no longer a person passing a document to another person, but an AI system carrying context across every interaction. The firms that win will be the ones that turn scattered internal knowledge into a live memory layer, then use it to automate decisions, coordination, and customer facing output faster than human teams can.