TL;DR:
- AI agents only make the quality of processes more visible instead of fixing them.
- Human workflows often depend on messy context and intuition that agents cannot reliably reconstruct.
- Agentic workflows need structured, machine-readable data, reliable interfaces, and strong governance.
- To make agents work, companies must redesign processes around agents instead of layering them onto old ones.
The magic bullet myth
Organizations tend to see AI as a magic bullet that helps them drive efficiency and reduce manual overhead. Yet, the operational backbone of most of these companies still relies heavily on human intuition, fragmented systems, and a complex web of ad-hoc, poorly maintained Excel files. Leadership often views AI agents as plug-and-play solutions: software you can simply “layer” over existing workflows to automate them and make them faster. They forget what AI fundamentally is: a technology with strengths and weaknesses that can only live on a strong foundation of data, compute, governance and culture.
Agents amplify broken workflows
Currently, many companies try to equip their employees with AI subscription licenses to automate and optimize processes. The idea is simple: use a low-code platform to build an agent that does the work for you. Too tempting is the thought of getting all benefits of AI just by booking a 50€/month subscription for some employees.
First attempts show that those agents can bring some benefit, but they alone are not enough to bring the productivity gains promised. Agents lack “human duct tape”: Humans can parse a shitty spreadsheet, ignore outdated tabs, and intuitively fill in missing context. AI agents cannot. As they take data literally, they are heavily reliant on setting their context correctly. Excel files are static, siloed, and often lack version control or programming interfaces. Agents require dynamic, real-time, and machine-readable data streams to function autonomously.
Pointing an autonomous agent at unstructured, inconsistent data doesn’t fix the process. It merely scales up the errors, leading to hallucinations and broken workflows.
What changes are required to make agentic workflows succeed?
The foundation agents require to work successfully is different from what humans can work with. Instead of only building new AI agents, the foundation is what companies have to focus on. And at its core, it’s fairly simple: structured, textual data in AI-readable formats like markdown with strong governance and tools instead of graphical user interfaces built for human eyes.
Agents require platforms, not files.
To act autonomously, an agent needs a sufficiently clean, governed, and structured data platform. It needs reliable interfaces and a single source of truth, not a shared folder of Excel files. For AI agents, we need an AI-readable stream of data instead of a file sent from A to B.
Transformation, not addition.
Adding an AI agent is not an incremental update, but a fundamental transformation. You cannot just add an agent to an existing process. Instead, you must redesign the process around the agent’s capabilities and constraints. As I wrote in The power of system design, questioning requirements and deleting parts can have a great impact.
Designing for machines, not humans.
Workflows must transition from producing human-readable reports (like highly formatted, color-coded Excel dashboards) to maintaining machine-readable environments (semantic layers, structured databases).
The platform is the engine, the agent just the steering wheel.
Ever since working hours peaked during the Industrial Revolution, we have spent the last 150 years successfully fighting for a shorter workweek. AI is the technology that could drive the next major leap forward. But this will not come without significant investments in the skills of people, data platforms, data governance, compute power and culture that lie at the foundation of the AI transformation. If we are able to master this basis, agents will be able to steer our processes and execute routine decisions entirely in the background, fundamentally changing the meaning of work.