Re-engineering 2.0: From Hammer’s Manifesto to Autonomous AI Workforces

In the early 1990s, Michael Hammer ignited a revolution with his seminal work on Business Process Re-engineering (BPR), urging organizations to radically rethink and redesign their processes to achieve dramatic improvements in performance, efficiency, and customer value.

Today, a new transformative force—Agentic Process Automation (APA)—builds on Hammer’s vision, propelling BPR into an unprecedented era of innovation.

Unlike traditional automation, APA leverages intelligent, autonomous agents that think, learn, and collaborate, acting not just as tools but as dynamic partners in orchestrating complex workflows.

APA – Agentic Process Automation

These agents anticipate challenges, optimize end-to-end processes, and unlock extraordinary value, echoing Hammer’s call for bold reinvention while harnessing cutting-edge technology.

Automating complex, thinking-based workflows needs new ways of analyzing work — very different from old-style rule-based automation. Traditional process mapping often falls short when dealing with AI agents that handle uncertain, flexible tasks.

Traditional Robotic Process Automation (RPA) uses process mining. This method pulls data from system logs (like ERP or CRM software) to map clear, repetitive tasks. It works well for structured work but has big limitations for cognitive automation.

Process mining misses the “white space” — things people do outside the system, such as reading emails, reviewing documents, or making judgment calls. It also requires deep technical integration and API access.

Agentic process discovery takes a better approach. It uses computer vision and machine learning to quietly watch how people actually work on their screens. It captures real task-level behavior without needing access to databases or complex IT setups.

This method removes human bias from process design and reveals patterns in messy, non-routine work. The data it collects helps train Large Action Models (LAMs), making it possible to automate the “long tail” of complicated business processes that traditional RPA could never handle.

Redesigning Workflows for Autonomous AI Agents

To use AI as a digital workforce, companies must redesign their processes around autonomous agents. Instead of rigid, step-by-step pipelines, workflows become dynamic and goal-oriented.

Key design patterns include:

• Planning Pattern (Interleaved Decomposition): The agent breaks big goals into small steps. It plans a bit, acts, checks the result, learns from it, and adjusts the next plan. This “plan-act-reflect-repeat” loop works well in uncertain situations and closely mimics how humans solve problems. For simpler, stable tasks, the agent can plan everything upfront.
• Multi-Agent Collaboration Pattern: For very complex work, a single AI model can fail or slow down. Instead, a team (swarm) of specialized agents works together. An orchestrator agent manages the overall goal and assigns subtasks to expert agents (e.g., research, coding, or compliance agents). This improves reliability, speed, and scalability.
• Reflection and Self-Correction Pattern: Before giving a final answer, the agent reviews its own work against company rules. It catches errors, unsupported claims, or hallucinations and fixes them. This self-check step is essential for high-stakes environments.

Human-in-the-Loop (HITL) and Deterministic Governance

Even with powerful autonomous AI, human oversight remains critical — especially in regulated industries like pharmaceuticals, finance, and manufacturing.

Human-in-the-Loop is not just a quick approval step. It must be carefully built into the workflow as a real control layer. In strict environments (such as those following EU Annex 22 or Digital GMP), AI must be predictable, traceable, and explainable. Purely probabilistic generative models are often wrapped in deterministic systems to meet these rules.

Tools like Temporal workflows help by turning unpredictable AI actions into reliable, stateful processes that can pause for human review before critical actions.

Good HITL design also uses smart interactions: the AI asks humans for extra information when its confidence is low. Human feedback then improves the AI over time. When done well — sometimes with augmented reality guidance — this collaboration greatly reduces errors in complex tasks.

In short, successful AI transformation requires fresh thinking about how work is discovered, designed, and governed — moving from rigid automation to flexible, intelligent, and responsibly supervised agentic systems.