Marketing knowledge layer

A “knowledge layer” for marketing is a shared store of facts an artificial intelligence system can look up before it writes. Shann Holmberg pitched that setup in an X thread, using campaign data, call transcripts and brand rules as the source material for marketing agents. (x.com) Holmberg’s post described one place to keep return on ad spend, headlines, calls to action, client notes and competitor information so an agent can draft ads and reports with the same inputs a human team would use. He also linked a GitHub starter so agencies could copy the structure and adapt it to their own accounts and clients. (x.com; docs.github.com) The underlying idea is retrieval-augmented generation, or RAG: instead of asking a model to rely only on its training, developers fetch relevant documents from a separate store and pass them into the response. OpenAI’s developer documentation describes that pattern as a way to answer questions from a company’s own content rather than from the model’s general memory. (developers.openai.com; developers.openai.com) In marketing, that separate store solves a familiar problem: performance data lives in ad platforms, brand rules sit in decks and documents, and sales or client context is buried in meeting transcripts. Google’s Marketing Analytics Jumpstart documentation describes the same fragmentation as a barrier to getting a “holistic view” of marketing performance across sources. (github.com) Holmberg has been building around that overlap between marketing operations and agents. A recent profile on GrowthFolks said he joined Espressio AI to lead agent product development after co-founding Lunar Strategy, a crypto-focused agency, and his public event listings on Luma include sessions on using AI agents in growth teams. (growthfolks.io; luma.com) The GitHub piece matters because a template turns a concept into something teams can fork, fill in and test. GitHub’s documentation says template repositories let users generate a new project with the same file structure and starter files, which is how many open-source playbooks spread. (docs.github.com; docs.github.com) The pitch also lines up with how agent builders are framing the stack in 2026. OpenAI’s agents materials describe built-in tools, file search and orchestration as the plumbing for systems that act on external context, not just chat from memory. (openai.com; developers.openai.com) What Holmberg added was a marketing-specific schema: not just documents, but the exact artifacts media buyers and creatives already use to make decisions. If agencies adopt that structure, the “knowledge layer” becomes less a buzzword than a filing system for teaching agents what converted, what matched the brand and what to write next. (x.com; developers.openai.com)