Let’s talk

RESEARCH / DIGITAL HUMAN

Coordinating work across applications

An architecture that separates planning, execution and control so agents can work across existing applications, with explicit goals, bounded actions and recorded handling of failures and exceptions.

Read on this page

At a glance

A business task rarely stays inside one application. An insurance request may arrive by email, require checks against a policy, create a claim record and end with a customer response. This paper describes an architecture for coordinating that work while keeping each agent's responsibility clear.

The Agent Execution Mesh separates the next goal, the actions needed to complete it and the control of the overall workflow. A worked claim-intake example spans Outlook and OpenKoda, including both successful processing and cases that should stop or take a different branch. For enterprise teams, the useful question is how a system handles verification, missing information and failure, as well as whether it can fill a form.

What the paper covers

  1. Planning, execution and workflow control are assigned separate responsibilities.

  2. Agents pursue bounded goals across existing applications using the page's semantic controls.

  3. The case study records exceptions and negative outcomes alongside successful actions.

Research context

The paper reports six executions of a specific claim-intake workflow, covering four terminal branches. This is a worked implementation with a limited execution sample; it does not establish reliability across arbitrary applications or all insurance processes.

THE ORIGINAL WHITEPAPER

Read it here.

Download PDF

The complete paper, with its original figures, tables and references.

30 pages · Read without leaving this page
FROM RESEARCH TO YOUR WORK

Bring your questions.

Discuss the approach, evidence and relevance to your team with the people building it.

Discuss this research