World Summit AI 2025: the era of agentic AI and large-scale production deployment

Agentic AI: from proof of concept to production deployment

The keyword of this edition: scale AI.

Large enterprises are no longer here to explore—they’re here to demonstrate. Kimberly-Clark, SAP, Adidas, BMW, Deutsche Bank, H&M, Zalando… all showcased their branded, orchestrated, fully operational internal AIs.

These companies share one common DNA: a unified, modular data foundation. Kimberly-Clark showed how this data layer enabled a 15% reduction in chemical waste and the deployment of business-oriented AIs such as Sophie (generative marketing content), a voice translation AI, and Dora, an internal assistant capable of leveraging the organisation’s collective knowledge.

SAP, meanwhile, presented Joule, an ecosystem of agents already covering 400 use cases—proof that agent platforms have become a new competitive arena. 

We have moved from talking about potential to demonstrating return on investment. 

Craig Stephen Slavtcheff - Chief R&D Officer, Kimberly-Clark

A critical Imperative: master Data before the next wave

A second major observation: companies are rushing to clean and structure their data ahead of the coming influx of next-generation information from physical and emotional sensors.

These new data streams (health, consumption, emotions) will redefine the relationship between brands and individuals by introducing a sensory dimension to data.

The challenge is enormous: build now an infrastructure able to ingest, analyse, and protect these highly sensitive flows.

Agentic AI: the rise of self-evolving systems and the emergence of ethical, responsible AI as a strategic imperative

Researchers emphasised one point: we are only at the beginning of the AI story. Artificial Intelligence is entering a phase of self-improvement—it analyses, corrects, and rewrites itself. This will shorten innovation cycles from months to days. The impact? A radical acceleration of technological pace and a fundamental question:

"How fast will AI evolve before we lose control?"

Companies are therefore urged to invest in explainability, ethics and continuous monitoring systems.

Notably, discussions no longer focus on AI bias, but on ethical and responsible AI. Projects now integrate transparency principles covering data, models and intended uses. What was once optional is becoming a strategic differentiator for large-scale deployments.

This evolution is not only technological—it is also regulatory.

The “ethical AI by design” approach aligns with the progressive enforcement of the EU AI Act, which requires more transparency regarding models used, data sources, risks, intended purposes, and control mechanisms. In other words: what used to be a best practice is now an obligation.

The objectives are twofold:

• protect users (especially in high-impact domains like healthcare, HR, finance, education, security),
• ensure auditability, traceability and explainability of systems.

A key point often overlooked was highlighted: these requirements also benefit companies.

In case of failure, error, or dispute, transparency enables understanding why a decision was made and allows quick correction.

And in a world where models evolve, rewrite themselves, or get replaced, traceability is essential—without it, regaining control or proving compliance becomes impossible.

In short: ethical, responsible AI is no longer a “nice to have,” nor a constraint. It is an accelerator for trust, compliance… and scale.

Deepfakes: the FBI’s warning

The closing keynote, delivered by the FBI, was a wake-up call.

The agency sounded the alarm about the proliferation of deepfakes—AI-fabricated audio, images, or videos impersonating real individuals—particularly those targeting teenagers. The phenomenon is already linked to a troubling rise in suicides in the United States.

15% of American teenagers have witnessed a classmate targeted by a deepfake.

Beyond psychological trauma, the consequences are wide-ranging: public humiliation, blackmail, extortion, and the spread of fabricated explicit content… Deepfakes can be created in minutes, shared in seconds, and are nearly impossible to erase once online. For a hyper-connected generation, the boundary between real life and digital identity is becoming dangerously thin.

But the threat extends far beyond individuals: our collective relationship with reality itself is at stake. When anyone can “see” a fake video of an influencer, artist, or politician, trust in visual evidence evaporates. People stop believing what they see—or worse, choose to believe only what confirms their opinions.

Between virtual influencers, remixed politicians, and generative avatars more realistic than humans, disinformation is entering a new era where proof and manipulation merge.

Agentic AI: a call for leadership and responsibility

This edition of the World Summit AI confirmed that AI is no longer an emerging technology. It is mature, deployed, and effective. But governance, ethics and international coordination remain immense challenges. Experts call for harmonised regulations to avoid a fragmented world split into competing technological blocs.

Use AI ethically for the good of humanity. Silence is not neutrality—it is complicity. 

Sarah Porter - Founder of the World Summit AI

Conclusion

World Summit AI 2025 marks the end of an era: that of exploratory curiosity.

We are entering the era of massive implementation, where every organisation must move from talk to action.

The leaders of tomorrow will be those who combine technological power with transparency and responsibility. AI is no longer just a promise—it is a mirror of our collective ability to use it wisely.