You have spent the past two years buying AI. New tools, new copilots, a few pilots that demoed well and then quietly underwhelmed in production. When that happens, the instinct is to blame the model or the vendor. The real cause is usually simpler and less comfortable: the AI cannot see what your company actually knows.
On June 12, 2026, Google Cloud published the Open Knowledge Format, or OKF, a small open specification for packaging organizational knowledge so that both people and AI agents can use it. It is not a product, and for most leaders the spec itself is not the point. The point is what it signals about where the value in AI is moving next, and it is worth understanding even if you never adopt the format.
The Bottleneck Was Never the Model
The knowledge an AI needs to be useful inside your business almost never lives inside the model. How your data is structured. What "active customer" actually means and which of three slightly different definitions is the right one. Why a process runs the way it does. Which report is authoritative and which is a stale copy someone forgot to delete.
That knowledge is scattered. Some sits in a data catalog behind a proprietary API. Some lives in a wiki nobody has updated since the last reorg. A lot of it lives in the heads of a few senior people who answer the same questions over and over. Every time a team builds an AI tool, it re-solves the same problem from scratch: gather that context, clean it up, and feed it to the model. The result is the confident wrong answer, the agent that quotes a metric defined three ways, the pilot that works in the demo because someone hand-fed it the missing pieces.
OKF is the first widely backed attempt to name that problem and give it a shared shape.
What Google Actually Released
OKF is deliberately unglamorous, which is its best feature. Knowledge is stored as a folder of Markdown files, each with a short block of structured tags at the top. There is no new software to run, no platform to buy, no account to create. A document describing a database table looks like a plain text file a person can read and an agent can parse, with links to the related concepts it depends on.
- sales/
- index.md
- tables/
- orders.md
- customers.md
- metrics/
- weekly_active_users.md
The idea traces back to a pattern AI researcher Andrej Karpathy described as the "LLM wiki." Humans abandon wikis because keeping them current is tedious bookkeeping. That tedium is exactly what AI is good at. An agent does not get bored, does not forget to update a cross-reference, and can revise fifteen files in one pass. So the format is built for people to set direction and agents to maintain the detail.
Two technical notes matter for context, and both connect to ground Honra has covered before. As an agent moves through an OKF folder, it reads a lightweight index first and only opens the deeper files it actually needs, which is the same efficiency principle behind progressive disclosure for AI agents. And OKF is not a competitor to the Model Context Protocol. MCP is the pipe that connects an agent to your systems. OKF is some of the content that flows through it. The two are designed to work together.
Why a Plain File Format Is a Strategy Signal
It is tempting to file this under engineering trivia. That would be a mistake. When a company the size of Google Cloud puts its name on a standard for packaging knowledge so agents can consume it, the signal is about where the constraint is moving.
A decade ago, the work that separated companies getting value from data and companies drowning in it was clean, well-governed data. Agent-ready knowledge is becoming the same kind of infrastructure. The organizations that pull ahead will not be the ones with the best model, because everyone rents the same handful of models. They will be the ones whose knowledge is captured, structured, current, and reachable, so that any tool they point at it gets smarter immediately. OKF is a marker that the industry now believes this layer is worth standardizing.
Read the Fine Print Before You Get Excited
OKF is version 0.1, and it earns some healthy skepticism. It standardizes the structure of knowledge, the folders and files and tags, but not the meaning. Two companies can both follow the format perfectly and still describe the same concept with different labels, so an agent fluent in one company's bundle gains little from another's. It is, as one fair critique put it, a shared way to store context, not yet a shared way to make sense of it.
It also carries Google's gravity despite the vendor-neutral framing. The reference tools that generate and serve OKF lean on Google's own models and data warehouse. The specification is young enough that Google's own software does not yet match its own written rules in every detail.
None of that makes OKF a bad idea. It makes it an early one. The right reading is not "adopt this format now." It is "the discipline underneath this format is where the durable advantage lives, and that discipline is true no matter which standard eventually wins."
What This Means for Leaders
The move here is not a purchase. It is a habit, and you can start it this quarter without committing to OKF or anyone else.
- Find out where your most important definitions live. Pick the five numbers your leadership team argues about most. If the authoritative definition of each exists only in one person's head or one analyst's saved query, that is a risk to your business and a ceiling on every AI tool you deploy.
- Treat institutional knowledge as an asset to maintain, not exhaust to ignore. The knowledge that makes an agent useful is the same knowledge that makes a new hire productive in week one instead of month three. The investment pays off with or without AI.
- Resist the platform reflex. The temptation when a trend like this lands is to buy a "knowledge platform" to solve it. Most of the value comes from the cheap, unglamorous discipline of writing things down, structuring them, and keeping them current. Spending your way past that step is a familiar and expensive mistake, the kind we wrote about in what business owners miss about technology overspending.
How Honra Reads This
At Honra we sit on the client's side of decisions like this one. We do not resell platforms or take a margin on the tools you adopt, so our read on OKF is the same one we would give you privately: it is an early and credible signal, not a product to rush into, and the discipline it points at is worth more than the format itself.
The work we care about is the work the spec does not do. Deciding which of your knowledge actually needs to be agent-ready and which is noise. Untangling the three conflicting definitions of your core metrics before you let an agent quote any of them. Building the system directly when that is the right call, and working alongside the team you already have when it is not. A file format will not do any of that. It just makes the case, in public and from a credible source, that the work is now worth doing.
Key Takeaway
The Open Knowledge Format is less a thing to adopt than a signal to act on. The next limit on what your AI can do for you is not the model. It is whether your own organization's knowledge is in a form a machine can use. Whether or not OKF is the standard that wins, the companies that treat their knowledge as infrastructure will get more out of every model they touch, and the ones that do not will keep wondering why the demos never survive contact with production.
