Return on Automation Return on Automation
EN

Area of expertise · 02

Agentic Automation

Automation is not a single tool, but a family of disciplines. The art lies in choosing the right tool for the right step — and in combining them into a coherent system. AI agents are the icing on the cake: powerful, but never the only tool.

The five disciplines I work with

Low-Code / No-Code

Applications and small processes assembled visually, often built directly by the business department. Strong when IT resources are scarce and quick solutions are needed — weak for enterprise-wide or highly regulated processes.

Workflow Automation

The backbone of the process landscape. End-to-end orchestration with an audit trail, ideal for structured, repeatable workflows such as approvals, onboarding or application processes.

Robotic Process Automation

Software robots that operate screens like a human. Strong where there is no API — as a bridge between systems that would otherwise be fed manually. Ideally a transitional technology, not the end state.

Intelligent Document Processing

Reading documents like a human — invoices, contracts, delivery notes. OCR, machine learning and large language models reliably extract even unstructured content.

Decision Management

Extracting business rules from code and making them editable for domain experts. Valuable for frequently changing or regulated rules — pricing, credit scoring, compliance.

AI agents — the icing on the cake

An agent knows the goal of a task and finds its own way there. It uses the tools mentioned above — starts a workflow, calls a decision engine, has an RPA bot do something, reads a document via IDP — and decides when to hand over to a human.

Important: an agent does not replace the other disciplines. It orchestrates them. Anyone who builds an AI agent without a clean workflow, without clear decision logic and without structured data is building on sand. That is exactly why a serious agent implementation starts with the fundamentals — and ends with the agent, not the other way round.

How I build agents

What agentic automation is not

It is no substitute for clean data structures, clear processes or good people. It is a tool that works well when the prerequisites are in place — and works poorly when it is hoped to paper over chronic problems. For me, this honest assessment is part of the craft: recognising when a use case is ready for an agent and when it is not.