If you’ve been experimenting with AI agents inside your marketing stack, you’ve probably run into the same frustrating pattern: the model is capable enough, but it keeps producing results that miss the mark. Wrong metrics, outdated definitions, context that just doesn’t line up with how your business actually works.
The easy assumption is that the model needs to be smarter. The harder truth is that the model is often just missing the right information to begin with.
Google Cloud just published something worth paying attention to here. They’ve introduced the Open Knowledge Format, or OKF, an open specification designed to solve what they call “the fragmented context landscape.” It’s worth understanding, not because you need to go build it yourself, but because it names a problem that directly affects every marketing decision your AI tools are trying to support.
The problem is organizational, not technical
In most organizations, the knowledge that AI agents actually need to do useful work is scattered across a dozen different places. Schema definitions live in a metadata catalog. Business definitions of key metrics live in a shared drive or someone’s head. Runbooks are in a wiki. Data lineage is buried in code comments or notebook cells.
When an AI agent needs to answer something like “what drove our engagement rate drop last quarter,” it has to assemble context from all of those incompatible sources. Every vendor solves this problem differently, with their own catalog, their own SDK, their own knowledge schema. Nothing is portable. Nothing talks to anything else without custom integration work.
The result, as Google Cloud describes it: “Every agent builder is solving the same context-assembly problem from scratch.”
As a CMO, this should matter to you because marketing generates and depends on a lot of that fragmented knowledge. What does “active user” mean in your context? What counts as a conversion? How is marketing-attributed revenue calculated versus finance-attributed revenue? These definitions exist somewhere in your organization, but probably not in a form that any AI agent can reliably find and use. Ask your head of sales, your head of marketing, and your controller what counts as a customer, and you will likely get three different answers; every agent you deploy inherits that argument.
What OKF actually proposes
OKF is deliberately simple. It represents knowledge as a directory of markdown files with YAML frontmatter. No new runtime. No required SDK. Just structured text files that live in version control, are readable by humans, and are parseable by AI agents without translation.
The fields are minimal: type, title, description, resource, tags, and timestamp. If you’ve ever used Obsidian or worked in a repo with a structured README convention, the shape will feel familiar. What OKF adds is a set of agreed-upon conventions so that a wiki built by one team can be consumed by a different agent without custom integration work.
The concept behind this draws from AI researcher Andrej Karpathy’s thinking on what he calls the LLM Wiki pattern. The core insight is that LLMs are actually good at the bookkeeping that causes humans to abandon internal wikis over time. They don’t get bored, they don’t forget to update cross-references, and they can touch many files in one pass. So rather than making an agent search scattered documents every time it needs an answer, you give it a shared markdown library that grows more useful over time.
Why this is a decision-making problem for CMOs
Here is where this lands for you practically.
The accuracy of any AI-assisted marketing analysis depends entirely on whether the model has reliable context about your business. If “weekly active users” means different things in your analytics warehouse than it does in your CRM, your AI tools will produce confident-sounding answers that are actually nonsense. If your attribution model has evolved but the documentation hasn’t, your agent is reasoning from stale assumptions.
This isn’t a model problem. It’s a knowledge management problem. And most marketing organizations don’t treat it like one.
The OKF specification is one proposed solution to this at the infrastructure level. But the more immediate implication for CMOs is this: before you invest in more AI capability, it’s worth asking whether your organization has done the work to make your business context legible to the tools you’re already using. That means clear, maintained, accessible definitions for your core metrics and data concepts, not just somewhere, but somewhere findable.
If your team can’t answer “where does a new agent go to learn how we define a qualified lead,” that gap is worth closing before adding another layer of AI on top.
OKF is early stage, v0.1, and this is a format proposal, not a finished product. But it points toward something real: the organizations that get the most from AI agents won’t necessarily be the ones with the most sophisticated models. They’ll be the ones whose knowledge is structured well enough for those models to actually use it.
Find out how to grow your marketing with the professional marketing minds at Ambient Array.
Original source: How the Open Knowledge Format can improve data sharing