TechEx Europe 2026: the next major AI breakthrough might not be in the model

What if the biggest mistake an AI system makes isn't caused by misunderstanding something, but by multiple systems each being individually correct?

That thought touches on one of the most fascinating developments in artificial intelligence.

Last week, TechEx Events Europe 2026 brought together an impressive community of international companies, technology experts and entrepreneurs in Amsterdam. From AI and cybersecurity to cloud infrastructure, robotics, data centres and digital transformation.

With organisations such as Google, Microsoft, IBM, Oracle, Snowflake, Shell, ING, Volvo Group, Deutsche Bank and Amazon Web Services (AWS) represented across the broader programme, it became clear just how many different disciplines are now contributing to the same technological future.

What an incredible and inspiring experience. I have tremendous admiration for the knowledge, ambition and openness of all the professionals I had the opportunity to listen to and speak with.

But perhaps the most interesting insights emerge when we stop looking at individual presentations in isolation and start exploring the connections between them.

The intelligence of an agent is not the intelligence of an organisation

Jaïro Martha from Snowflake addressed a fundamental challenge within autonomous business processes in "Every Agent Was Right. The Business Was Wrong."

Imagine a marketing agent offering customers next-day delivery. A sales agent confirms the orders. Meanwhile, a supply chain agent has already reserved the same inventory.

Each agent acts logically based on its own information and responsibilities. Yet together, they produce a business outcome that nobody intended.

This is not a traditional AI problem where a model generates an incorrect answer. It is a coordination problem between independently operating systems.

This idea connects closely with "From Data Lakes to Data Products – What's Next?", featuring Dik Harbrecht MSc. from Shell, Ivan Branco from Volvo Group and Aatish Kumar from Eneco, among others.

Data products make information reusable while bringing ownership, quality and definitions closer to their source. But even an exceptionally well-organised data landscape cannot guarantee correct collective decision-making.

An inventory position may be accurate at the moment it is retrieved, while another agent reserves that same inventory a second later.

Shared information is not the same as shared decision-making.

I believe this reveals the next architectural challenge: agents must not only understand the same business context, but also account for each other's ongoing actions, commitments and the consequences of their decisions.

That requires more than a central data source. It calls for consistent business rules, up-to-date transaction states, reservation mechanisms and explicit agreements about which system has decision-making authority, and when.

For years, we have developed systems that optimise their own individual processes. Agentic AI now forces us to consider how those optimisations relate to the interests of the organisation as a whole.

Autonomy begins where technical access meets business accountability

This complexity increases even further when AI is allowed to act independently.

During "From Assistants to Autonomy – Architectures Enabling Next-Gen Agents", architectural challenges involving planning, reasoning, tool orchestration, memory and human oversight were central to the discussion featuring Avinash Jha from Volvo Group, Frank J. Kartmann from Deutsche Bank, Anna van den Breemer-Kleene from EY and Csaba Tamas from Parloa.

What makes this especially interesting is how it connects with two other sessions.

Constantine Grigoroff from Shell explored the use of MCP, semantic metadata, security policies and evaluation methods to connect agents reliably with enterprise systems in "Beyond the Demo: Building Trusted AI Agents with MCP."

Stefan Malovic from Okta, in "Digital Co-workers, Real Risk", focused on agent identity, access rights and accountability.

Together, these sessions point towards a crucial distinction:

Just because an agent is technically capable of performing an action does not mean it is authorised to perform that action within the scope of its original assignment.

An agent might have permission to modify customer information. But should it be allowed to change contractual terms independently? And does its original mandate remain intact when it delegates a task to another agent?

My conviction is that autonomy should not be restricted solely by system access, but also by intent, context and potential consequences.

An agent should be able to demonstrate not only who it is and which tool it uses, but also on whose behalf it acts, why that action is justified and within which boundaries it is permitted.

This transforms the challenge from access management into something much broader: how do we design verifiable digital accountability?

And that becomes essential when multiple agents begin collaborating not only with employees, but also directly with agents operating across different organisations.

A successful AI implementation is not a delivery milestone, but a continuously governed business process

This is where a third fascinating connection emerges.

Csaba Tamas from Parloa explored why increasingly intelligent models do not automatically translate into reliable AI agents in production in "Smarter Models, Same Failures."

After all, the reliability of an entire workflow depends on far more than the model itself: external systems, exceptions, recovery mechanisms, monitoring and the ability to correct errors quickly.

Charlie Doubek from NTT DATA approached a related challenge from a different perspective in "Scaling AI: Why Architecture and Operating Models Matter More Than Perfect Data": organisations should not wait for perfect conditions, but neither should they remain trapped in isolated pilots without scalable architecture and clear responsibilities.

And Koen van Muijen from ING Netherlands, in "Boosting Product-Led Thinking with ING NL", explored product ownership, strategic alignment and reducing organisational dependencies.

I find this combination particularly valuable.

Because who manages an autonomous agent after it has been successfully implemented?

The developer who built the software? The IT department responsible for the infrastructure? Or the process owner accountable for the actual business outcome?

If an agent is allowed to create value autonomously, ownership of its performance, risks and continuous improvement must be equally explicit.

That is why I find it more meaningful to evaluate AI based on realised business value rather than the percentage of tasks completed without human intervention.

An agent that independently handles 90% of customer enquiries is not automatically better than one handling 70%, if the latter demonstrably makes fewer mistakes, achieves better customer outcomes and requires less corrective work.

The quality of autonomy is not determined by how much control we remove, but by how much responsibility we can reliably delegate.

The next step requires more than intelligent technology

When I connect these different perspectives, a fascinating development begins to emerge.

The challenge is shifting from developing individually intelligent AI applications towards building cohesive, scalable ecosystems in which information, autonomy, authority and accountability are aligned.

That is also the direction we are pursuing at AVENDOX.

Not automating simply because it is technically possible, but developing systems that collectively deliver demonstrably better results. Where autonomy is not an end in itself, but a carefully designed means to an end.

Perhaps the most competitive organisation of tomorrow will not be the one with the most AI agents.

But the one that knows how to orchestrate human and artificial intelligence most effectively.

TechEx Europe was a wonderful demonstration of what becomes possible when experts from different disciplines bring together their knowledge, ambitions and perspectives.

My appreciation goes out to all the speakers, organisations and professionals who share their expertise and continue to advance these developments. The presentations and conversations once again highlighted just how many fascinating challenges still lie ahead.

Technology is evolving at an extraordinary pace. Real progress happens when the way we organise, decide and collaborate evolves alongside it.