EKA in Practice: What Happened When We Deployed the Framework

What we learned by deploying Executable Knowledge Architecture in practice, including where the framework held up and what changed.

We’ve deployed Executable Knowledge Architecture at a professional services firm with significant liability exposure. The implementation details are competitive advantage and can’t be shared publicly, but the pattern is transferable.

Orientation, Controls, and Methodology Records

New users experience the oracle pattern’s failure mode firsthand by role playing during the orientation. They are prompted to ask naive questions and receive plausible (and possibly correct!) answers but then discuss how those answers cannot be verified without reproducing the effort to obtain them independently. By the time they reach the assistant pattern role play section, they understand the deficiencies of the oracle model. Description of a policy is far less effective a learning tool than seeing the failure mode in action.

The tooling itself distinguishes between exploratory work (oracle pattern, disposable) and analytical work (assistant pattern, code required). The governance model mandates that the system require executable artifacts.

Client confidentiality requires us to isolate projects. Each project has its own directory and AI context, preventing information from one client entering another client’s work and minimizing legal exposure from accidental disclosure.

We version and document every analytical artifact, including the reasoning behind it. When someone asks “how did you calculate this?”, as they will in our industry, we can show the code implementing the expert’s specified method.

Adoption and Reuse

We launched a month ago. In three weeks, over Christmas, 33% of eligible professional staff across North America used the system to produce at least one client-facing artifact. Adoption exceeded what we’d imagined possible. We’d expected resistance, but skeptical professionals wanted to use the system once they understood the distinction. The assistant pattern gave them something the oracle pattern could not: confidence that their judgment remained central and that the tool extended their capabilities without replacing their accountability.

Professionals use one another’s workflows as templates and turn repeated methods into reusable tools. After only a month, we’re seeing the system’s value grow faster than the number of users because each can reuse others’ work.

Early drafts of the framework treated oracle and assistant as competing approaches. In practice, people use both: oracle for ideation, brainstorming, and exploring a problem; assistant for work that must be reproducible, defensible, or delivered. Using the system has made the distinction clearer than describing it in theory did.

Some professionals object: “not everything needs code,” “my expertise is ineffable,” “this is slower than just asking.” Their objections tell us how they understand their own value. Discussing those beliefs is often more useful than the training itself.

Client Value and Professional Judgment

The framework works imperfectly but well enough that adoption is accelerating without our prompting.

Clients pay for the professional’s methodology in an inspectable, versioned, reproducible form. The governance requirements make that deliverable possible. AI speeds up encoding the method while professionals retain responsibility for the judgment.

The firms that figure this out will have something their competitors don’t: a way to use AI that increases rather than undermines their professional value. The ones that treat AI as an answer machine will face the two-prong threat described elsewhere on this site: malpractice risk if they trust it, redundant effort if they verify it.

What’s Next

We’re building shared repositories of validated workflows and collecting session data to study how professionals use the system. Recording methodology also preserves expertise through personnel changes, giving us a use for the artifacts beyond auditing.

If you’re deploying AI where professional accountability matters, I’d like to compare experiences. I suspect we face similar challenges. The implementation details remain private, but other firms can use the pattern.

Email mike@molsen.ca.