One Definition, Every Surface

October 2, 2026 · Akoonu Team

You will have more than one agent surface. Maybe not today, but someday soon.

Someone’s using Claude with Salesforce through MCP. Someone building an Agentforce agent. There’s a prompt template in Prompt Builder. A Flow calls a model. A developer wrote something in Apex.

Each of those is a place your definitions live. And each one is a place they can diverge.

How drift happens

Nobody decides to have four definitions of “open pipeline.” It happens one surface at a time. The Agentforce builder writes the field list into a topic. The MCP user gets whatever the native server returns. The Flow has its own SOQL. Six weeks later, a stage is renamed, one of the four gets updated, and nobody can say which number is right in the Monday meeting.

The failure is quiet. Nothing errors. The answers just stop agreeing.

One place, every surface

ContextWorks holds the definitions in one place — the Salesforce org, which is the system of record — and delivers them to every surface from there.

  • MCP clients (Claude, ChatGPT, anything that speaks MCP) call ContextWorks tools and get the curated entity.
  • Agentforce reaches the same entities through actions.
  • Prompt Builder and Flow pull the rendered context as text.
  • Apex gets a public API.

It works because ContextWorks never calls a model. It returns markdown or JSON. Anything that reads text reads the context, which means there’s nothing to integrate per vendor and nothing that breaks when you switch.

”One definition” means one place, not one payload

Worth being precise here. The claim isn’t that every agent receives identical context. Domains scope what each agent reaches, and instructions differ by entity, field, and element — the Pipeline domain and the Renewals domain carry different guidance on the same amount field, on purpose.

The claim is that context is managed in one place and delivered to every agent. Change a definition once and every surface that reads it gets the change. That’s what “change a definition, not a prompt” means in practice.

Why it matters beyond consistency

Nobody knows which AI tools they’ll be using in two years. Context built inside one agent platform is stranded there. Context built in the system of record outlives whatever you connect to it.

See ContextWorks in action.

A short walkthrough of how context gets built, what an agent receives, and how it reaches the tools your team already uses.

Trusted by revenue teams at

HoneywellThalesAvery DennisonTeamworksKOREStreetLight Data

We use cookies to understand how you use our site and improve your experience.