4 waves. 1 pattern. And why the era of ‘Agentic AI’ demands a human foundation.

Over the past 20 years, I have witnessed four major waves of transformation emerging from the engine room: Agile, Digital, Generative AI – and now Agentic AI.

Each wave brought new buzzwords, more expensive consultants and more spectacular demos. Yet the phenomenon behind it all has always remained the same: 👉 Companies buy the technology, skip over the culture, and wonder why the transformation fails.

We’re seeing it again today: in 2026, tech conglomerates will be sacking tens of thousands of employees in the name of imagined ‘AI productivity’ – whilst studies show that only 2 per cent of AI initiatives deliver genuine, measurable business value. They are discarding tacit knowledge and unleashing unprotected systems onto chaotic data graveyards.

I have set out my experiences, lessons learnt and frameworks from recent years in a series of books. They are not high-flown theoretical tomes, but practical tools.

Here is the essence of this journey, distilled into four pillars:

FlowOS: People before processes (The end of the ‘agility’ charade)

Agility was meant to set us free, but has often merely brought us more meetings, SAFe bureaucracy and micromanagement in sprint guise.

The premise: Transformation must free up time and cognitive capacity – otherwise it will be rejected.
The symbiosis: FlowOS combines Design Thinking (empathy & Discovery Gate), OKRs (strategic purpose through two-level goals) and Kanban (visibility & flow).
In practice: No more five daily stand-ups that degenerate into coffee klatches. Instead, one weekly focus sync, four-week cycles and radical protection for genuine deep work.

OrgDNA: The constitution for the human-AI organisation

Before a single AI tool is introduced, an organisation doesn’t need a 50-page policy document, but rather a cognitive constitution:

The 4 guiding questions: Who are we? What are our non-negotiables? Who decides what? What do we protect?
Psychological safety instead of fear: People don’t resist AI out of ignorance, but because they feel their identity is under threat. Those who ‘put the pen in their hands’ (clear accountability) and enable them to ‘visit the cockpit’ (transparency) transform fear into curiosity.
Data Trust Card: If vectors and embeddings make all data uniform, we need visible provenance, expiry dates and access rules for every data pool.

Human Before the Loop (HB4L) & Agentic Architecture

Traditional Human-in-the-Loop (HITL) is often a dangerous fig leaf: when people have to click ‘Approve’ 400 times a day, approval fatigue sets in – they become blind rubber-stamp robots.

Human Before the Loop: The crucial human work takes place before the agent starts – context design, rights configuration, error architecture and clear naming responsibilities in the agent registry.
No prompts as a safety fence: A language model does not understand ‘Please be honest’. Security must be deterministically embedded in actual code!
- MCP (Model Context Protocol): The universal socket for data and tools.
- ACP (Agentic Control Protocol): The incorruptible shield & safeguard that hard-codes the blocking of unauthorised actions.
- Human-on-the-Loop (HONL): Humans as conductors in the control tower for genuine exceptions (bona fide), whilst mala fide attacks are technically halted.

Multivendor swarms & European sovereignty

The future does not belong to the company with the one ‘best model’, but to the one that can confidently orchestrate different models:

Model diversity: Mistral for reasoning, Aleph Alpha for domain knowledge, specialised open-source models for bulk tasks.
Protection against the CLOUD Act & FISA 702: On-the-fly anonymisation gateways before data leaves European cloud borders.
Countering the asymmetry of swarm attacks: Attackers never rest. Through constant AI-assisted red teaming and ‘digital profilers’, we continuously harden our own systems.

The conclusion: From hand sawyer to windmill operator

In the 16th century, hand sawyers in Holland feared the wind-powered sawmill – yet the mills turned the Netherlands into a leading seafaring nation and created entirely new, demanding professions.

AI does not destroy work; it destroys the illusion that work must never change.

Anyone who introduces AI to cut people loose is building an unmanageable time bomb. Anyone who introduces AI to turn people into conductors, designers and guardians of values is building the sustainable organisation of tomorrow.

Human Before the Loop. Always.

Here are the books, which I have published in cooperation with different AI assistants.