Speaking at Cannes Lions 2025, Microsoft AI CEO and DeepMind co-founder Mustafa Suleyman described a turning point in how we relate to AI. “We’re in this moment… where that intelligence is about to be set free from these objects and to travel with us as a companion.” AI, in his framing, is no longer just a technical layer. It’s becoming something creative teams engage with throughout the process—reading context, adapting tone, and collaborating in real time.
From Games to Real-World Creativity
Suleyman recalled that much of the early progress in AI happened in game-like environments because real-world applications weren't yet possible. In 2012, he watched one of the first generative models produce a new handwritten digit. “Suddenly the number seven appeared.” That moment revealed the core idea: human creativity is built on recognising patterns in complex spaces, and those patterns can be modelled computationally.
Now, the systems have scaled. “It would be the equivalent of every single person on the planet holding a calculator, doing a complex sum every second, every day, for an entire year—for 400 million years.” But computation alone isn’t the point. The power lies in how these models are being used to interpret language, respond with tone, and take initiative.
Emotional Intelligence
“People were very skeptical that EQ was going to work,” Suleyman said, referring to emotional intelligence. But the models have evolved. “As the models get larger and more capable, they get easier to steer.” His team focused early on the tone of responses: “How do we introduce kindness… how do we have it acknowledge and respect and recognise what you’ve asked?”
That attention to tone is now part of how brand messaging is being scaled across campaigns, especially when assistants and co-pilots are being trained to reflect brand voice.
Creativity and the Value of Hallucination
When asked about hallucinations—AI producing false or made-up content—Suleyman pushed back on the framing. “We want that ambiguity… we want the fuzziness.” He sees it as an engine for lateral thinking. In contrast to traditional software, which retrieves what’s already known, generative AI creates something not yet defined. The result isn’t always accurate, but often creatively useful.
This idea of “unexpected collisions of thought” aligns with how many creatives develop early-stage concepts—starting from imperfect inputs and building from there.
The Rise of Agentic AI
Suleyman mapped out three phases of AI evolution: recognition (e.g. transcription, detection), generation (e.g. creating content), and now, action. The next wave of models will be built to execute complex tasks—not just generate ideas but carry them out over time. “Actually build a project or run a plan out,” as he put it.
This shift from input–output interactions to ongoing systems will reshape creative operations. Teams will move from using AI to brainstorm or draft, to using it to schedule, deploy, and iterate in production environments.
Access Is No Longer the Advantage
There’s little gap now between proprietary and open-source models. “The APIs are becoming commoditised and radically cheap,” Suleyman said. “Many of the models are basically open source now.” The consequence: anyone can access advanced capabilities. Advantage will come not from owning the tools, but from knowing how to use them well.
He dismissed the idea that AI’s future will be monopolised. “It just doesn’t resonate,” he said, referencing the outdated belief in a singularity dominated by one or two players.
Standards Are Rising
As more teams start using AI, the volume of content will increase—but so will expectations. “The quality bar is about to go through the roof,” Suleyman said. “Judgment is still going to matter. Curation is still going to matter. Brand is going to matter more than ever before.”
Creativity won’t be flattened. It’ll be tested—by abundance, by automation, and by audiences who expect more coherence and cultural sharpness than ever.