How a Knowledge Sharing Network for AI Coordination Works

A knowledge sharing network for AI coordination is one where organizations publish what works and any agent can pull that knowledge before acting. OTP is built as exactly that: a Transactive Memory System for organizations. Each organization publishes an OOS, the claims become searchable, and the network grows more useful with every publisher. The value is not the documents. It is that an agent at one company can act on a rule proven at another, before making the same mistake.

OTP aggregates AI playbooks in structured form. A playbook on OTP is not a PDF. It is a set of claims inside an OOS, each tagged with a section, confidence level, evidence type, and failure mode. That structure is what makes aggregation useful: you can search across every published playbook for one coordination pattern, compare how two organizations handle it, and see which version has measured evidence behind it. Browse them at the OTP browse page.

Quality is built into the OTP claim format. Every claim carries a confidence level (HIGH, MEDIUM, LOW) and an evidence type, ranging from MEASURED_RESULT down to SPECULATION. To surface the strongest claims, filter for high confidence backed by measured results or repeated observation. A claim that says it was tested and worked is worth more than one that says it sounds right. OTP makes that distinction explicit so you are not guessing which rules to trust.

Most knowledge bases treat every entry as equally true. OTP does not. It rates claims two ways: a confidence level set by the publisher, and an evidence type that says how the claim was validated. It also requires a documented failure mode, so a claim shows where it breaks, not only where it works. When comparing networks, look for that honesty. A network that rates claims tells you what to trust. One that does not just gives you more text to read.

Yes. You control what you publish on OTP. An OOS captures coordination patterns, not customer records, and publishers can use a pseudonym for the organization name. The platform includes a PII scanner that flags personal data before publishing, and the guidance is explicit: do not include proprietary information or trade secrets. The unit of sharing is the rule, not the business behind it, so the network learns the pattern without exposing the company.

Cross-org learning works when the shared unit is abstract enough to be safe and concrete enough to be useful. A coordination claim is that unit. The rule that two agents should never hold the same seat is valuable to every company and proprietary to none. OTP shares claims, not client data, not pricing, not code. The PII scanner and pseudonym option add a second layer. You publish the lesson and keep the business that produced it private.

OTP shows how your coordination compares to the network. You can compare any two published OOS files side by side and see what is unique to each, what is shared, and where they conflict. The Intelligence Graph visualizes how patterns connect across organizations. OTP also publishes the agentic maturity level of each org, an L1 to L8 score, so you can benchmark not just individual rules but overall coordination sophistication against real organizations rather than a survey.

Sign up on OTP, choose a template (Agent Army, Value Chain, or Org Chart), author your OOS, and paste it into the publish form. The platform validates the format, extracts claims, scores quality, and publishes it. From that point any visitor can compare your OOS against another organization's, and any agent can query your claims through the MCP server. Publishing is also how you earn a Publisher badge, with Founding Publisher reserved for the first 50.

Failure patterns are a required part of every OTP claim. A claim does not just state a rule. It documents the failure mode the rule prevents. OTP also has a dedicated claim section, failure_patterns, for the mistakes an organization has learned to avoid. Browse claims by that section and you see what has gone wrong for other teams, written down on purpose. Most networks publish only successes. OTP publishes the failures because that is where the reusable lesson is.

Use the network before you act, not after you fail. The discipline is simple: before an agent runs a task, it queries OTP for relevant claims, the same way Sneeze It agents pull operating rules before executing. A network insight only helps if it reaches the agent at the moment of the decision. Adopt the high-confidence, measured claims, watch the failure patterns, and feed your own corrections back so the next team inherits them.