Definition
The three builders are the composition tools used to assemble production AI solutions: workflows, which define process logic; agents, which handle autonomous execution as atomic, model-agnostic components; and apps, which surface the moments where human judgment is required.
What are the three builders?
The three builders are the composition tools used to assemble production AI solutions on a horizontal AI platform: workflows, agents, and apps. Each addresses a distinct layer of how an enterprise AI solution is constructed, executed, and maintained.
Think of a production AI solution as having three moving parts. Something has to define the process and its logic. Something has to execute the autonomous work. And something has to bring humans in at the moments where judgment is still required. The three builders map directly to these three requirements, making it possible to compose a complete AI solution from reusable components rather than hand-coding each from scratch.
What does each builder do?
Workflows define what needs to happen: the process logic, sequence of steps, and decision points. They are built to be malleable, so the process can evolve without requiring the entire solution to be rebuilt.
Agents handle what can run autonomously. Each agent is an atomic, self-contained component built around a specific task. Because agents are independent units, they can be swapped, upgraded, or replaced as models improve without touching the surrounding workflow. Agents are also model-agnostic by design: each can run on whichever model best suits the task, and switch to a better or cheaper option as they become available.
Apps surface the moments where human judgment is still required. They are the interfaces through which people approve decisions, handle exceptions, and stay connected to the work AI is doing on their behalf.
Why does separating the three builders matter?
When process logic, autonomous work, and human interfaces are conflated into a single system, every change to one requires touching all three. The three-builder model separates these concerns so each layer can evolve independently. A workflow that needs a new approval step does not require changes to the agent layer. An agent that needs to be upgraded does not require changes to the application. This is what makes assembly-first architecture composable and maintainable at scale.
Can a single use case utilise all three builders simultaneously?
Yes, and most production enterprise AI solutions do. A use case might use a workflow to define the overall process, multiple agents to handle autonomous steps within it, and an app to surface the points where a human needs to intervene or approve.
How do the three builders relate to vertical decoupling?
Vertical decoupling is the principle; the three builders are its implementation. Vertical decoupling separates what needs to be done from who or what does it. Workflows capture the what, agents handle the autonomous execution, and apps bring in the human at the right moment.
What happens when enterprises build AI without separating these three layers?
Solutions become rigid and expensive to maintain. Upgrading an agent requires rebuilding the application around it. Changing a process step requires touching the integration layer. The cost of change scales with the size of the deployment, which is exactly when flexibility matters most.
Are the three builders relevant to all types of enterprise AI use cases?
Yes. The specific workflows, agents, and apps differ across Finance, HR, Operations, and other functions, but the structural requirement for all three layers is consistent across use cases.
