A Process Harness for Uplifting Legacy Workflows to Agentic BPM: Design and Realization in CUGA FLO

2026-06-25 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Expert, medium

Summary

Fabiana Fournier and Lior Limonad introduce the process harness, a novel mechanism designed to uplift legacy workflows into Agentic Business Process Management (Agentic BPM) without requiring replacement of the underlying deterministic workflow engine. This harness integrates a policy-governed agentic layer around the existing engine, intercepting designated control points to provide reasoning, adaptation, and oversight, while the engine maintains structural authority. The authors develop the Task-Decision-Flow (TDF) model, which specifies both data schema and execution semantics, decomposing LLM reasoning across three agent types: a TaskAgent for knowledge-intensive execution, a DecisionAgent for per-case gateway routing, and a FlowAgent for runtime flow adaptation via a principled hook mechanism. Each agent's reasoning is guided by explicit policies from the process FRAME. CUGA FLO serves as the design and implementation of the TDF model, demonstrated effectively on a loan approval workflow that utilizes all three agent types and hook-driven regulatory override. This approach uniquely reconciles imperative structural compliance with normative agentic autonomy.

Key takeaway

For AI Architects or Automation Engineers seeking to modernize legacy business processes, you should consider implementing a process harness to introduce agentic capabilities without a full system overhaul. This approach allows you to enhance existing deterministic workflows with LLM-driven reasoning and adaptive oversight at critical control points, ensuring both structural compliance and flexible, policy-governed autonomy. Evaluate CUGA FLO as a reference implementation for integrating Task, Decision, and Flow Agents into your current BPM systems.

Key insights

A process harness integrates policy-governed agents into legacy workflows, adding reasoning and adaptation without replacing the core engine.

Principles

• Agentic layer provides reasoning and oversight at control points.
• TDF model decomposes LLM reasoning into specialized agents.
• Policies from process FRAME govern agent behavior.

Method

The Task-Decision-Flow (TDF) model defines data schema and execution semantics, using TaskAgent, DecisionAgent, and FlowAgent types for LLM reasoning, gateway routing, and runtime flow adaptation, respectively, all governed by explicit policies.

In practice

• Apply a process harness to existing deterministic workflow engines.
• Implement TaskAgent for knowledge-intensive tasks.
• Use FlowAgent for runtime flow adaptation via hooks.

Topics

• Agentic BPM
• Legacy Workflow Uplifting
• Task-Decision-Flow Model
• LLM Agents
• Business Process Management
• CUGA FLO

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, AI Architect, Automation Engineer

Related on AIssential

• Blueprint First, Model Second: A Framework for Deterministic LLM Workflow — Decoupling LLM decision-making from workflow execution via code blueprints ensures deterministic, verifiable agent behavior.
• Adaptive Auto-Harness: Sustained Self-Improvement for Agentic System Deployment on Open-Ended Task Streams — Adaptive Auto-Harness enables LLM agents to sustain performance on dynamic, open-ended task streams by continuously adapting their operational harnesses.
• A Formal Framework for Declarative Agentic AI in Business Process Analysis — A formal framework using AGO methodology and a BPKB enables precise, automated, and sound Agentic AI for business process analysis.
• A harness for every task: dynamic workflows in Claude Code — Claude Code's dynamic workflows enable on-the-fly multi-agent orchestration to overcome single-context AI limitations.
• not much happened today — AI agentic workflows are maturing with advisor-executor patterns, robust harnesses, and more realistic evaluation benchmarks.
• Bridging intent and execution in agentic systems — Minimizing the intent-execution gap in agent harnesses is crucial for state-of-the-art LLM performance.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.