Talks about the architecture behind useful AI

When AI begins to act, the difficult questions reach beyond the model. Who authorises a commitment? How do business rules reach the system that executes it? What does a successful demonstration actually prove?

I explore these questions through enterprise architecture, systems thinking and independent experiments. These talk concepts can be adapted for leadership briefings, architecture communities, conferences and moderated discussions in English or German. Scope and timing are agreed for the event.

Governed autonomy: who may let the agent act?

An agent can find a useful option before the company has decided who may approve it. This talk follows that gap from a fictional cross-department case into concrete architecture questions: bounded permissions, authoritative records, approval at the point of action and recovery after partial failure.

For leaders, architects and governance teams. Participants take away a set of questions for deciding where autonomy belongs and where responsibility must remain explicit.

Enterprise architecture becomes operational

A business rule earns its place in an architecture when changing it changes what the systems permit. Starting with my Enterprise Agentic Mesh proposal, this talk explores how approved business definitions can reach versioned checks, permissions and coordination across existing software, people and agents.

For enterprise and solution architects, technology leaders and operating-model teams. Participants take away a way to trace one business decision to its implementation and test whether that translation works. The proposal and fictional examples are labelled throughout.

Evaluate agents beyond the demo

A convincing result may conceal lost evidence, exhausted budgets or repeated actions. My graph-walk and character experiments expose a more useful question: which mechanism worked, under which conditions, and what still needs testing?

For architects, engineering leaders and teams evaluating agent pilots. Participants take away a distinction between mechanism tests, held-out evaluation and operational qualification, plus questions for judging stopping, failure and recovery.

Invite a talk or conversation

Share the audience, event context, preferred language and the question you want participants to leave better equipped to answer. Discuss a speaking invitation.

Timotheos Samartzidis is a systems thinker and enterprise architect based in Wiesbaden, Germany. He connects AI strategy, governance and systems design. His background spans enterprise architecture, regulated product engineering and research in computer vision and biometrics. Independent views do not represent his employer.