Takeaways from the AI Summit London: from experimentation to scalable deployments

I attended the 10th anniversary of the AI Summit, which took place in London on 10^th and 11^th June 2025, and wanted to share some highlights and takeaways

The event marked a clear inflection point for the industry. With over 5,000 attendees and more than 400 speakers, the event reflected how AI has moved beyond experimentation into real-world deployment. 

The conversation has shifted accordingly. The question is no longer just about what AI can do, but where it should be applied, how it can scale, and what organisational changes are required to extract value. Across sessions, the message was consistent: AI is already reshaping industries, but the next phase depends on execution rather than capability.

New possibilities: rapidly expanding capabilities

The most striking aspect of the summit was the breadth of new use cases emerging simultaneously, many of which were inconceivable even two to three years ago.

One clear signal is AI’s move into the physical world. Robotics is no longer confined to controlled environments; it is operating at scale in cities. Delivery robots such as Coco are now navigating pavements, handling adverse weather, and integrating into existing urban infrastructure. What stood out from both sessions and floor discussions is how quickly this is evolving. The next step—already visible in early demonstrations—is the progression towards more advanced, humanoid-like systems, such as Ameca from Engineered Arts, capable of interacting more naturally with environments and people. This moves AI beyond automation into something closer to embodied intelligence, with implications for logistics, retail and public services. 

At the same time, enterprise AI is undergoing a structural shift. The move towards agentic systems is redefining how work is organised. Instead of isolated tools, organisations are deploying multiple agents across workflows—customer care, finance, HR—capable of planning, executing and iterating tasks. The result is a transition from human-led processes to AI-orchestrated workflows, with humans moving into supervisory and decision-making roles.

In consumer-facing environments, this is translating into what can best be described as “AI concierge” models. In retail, for example, Pandora highlighted how conversational agents are beginning to replicate the contextual interaction of in-store experiences—understanding intent, guiding decisions, and increasingly engaging proactively rather than reactively. The user interface shifts from transactional (search and filter) to conversational, continuous engagement.

Creative industries provide another illustration of how far capabilities have advanced. AI is compressing production cycles dramatically, enabling feature-length films, such as Hell Grind, to be produced in weeks rather than years and at a fraction of traditional cost. This does not just improve efficiency; it changes who can create. Lower barriers to entry are enabling new participants to produce high-quality content, particularly in areas such as VFX-heavy storytelling.

Taken together, these examples point to a broader shift. AI is no longer just improving existing processes; it is enabling entirely new ones. The opportunity lies less in incremental gains and more in rethinking how industries operate when intelligence, automation and content creation are no longer constrained in the same way.

Challenges: what stands in the way

For all the progress, the summit also highlighted a clear set of constraints that will determine whether these possibilities translate into real outcomes.

Governance and trust emerged as the most immediate challenge. As AI systems become more autonomous and more embedded in real-world processes, the question of control becomes central. In practice, this is less about restricting AI and more about ensuring visibility—understanding how systems behave, when they fail, and how decisions are made. Without this, large-scale deployment—particularly in regulated sectors—will remain limited. 

Fragmentation is another structural issue. Differences in regulation, lack of standardisation, and siloed enterprise systems are creating friction across the ecosystem. This is particularly evident when scaling beyond pilots, where integration across functions and geographies becomes critical.

Data remains a persistent bottleneck. Despite advances in models, outcomes are still constrained by access to high-quality, well-structured data. Organisations continue to struggle with fragmented systems and security constraints, limiting their ability to deploy AI effectively at scale. The issue is not model capability, but enterprise readiness.

Perhaps the most important constraint, however, is organisational. A recurring theme was the gap between individual and organisational readiness. Employees are increasingly comfortable using AI tools, but companies are not redesigning workflows to take advantage of them. This leads to fragmented adoption, shadow AI, and limited impact. The underlying issue is that AI is often layered on top of existing processes rather than used to reimagine them.

Finally, there is a risk of misapplication. Not every use case requires AI, and in many cases, deterministic automation provides greater reliability. Overusing AI where judgment is not required, i.e. just using AI for the sake of using it, introduces unnecessary complexity without improving outcomes.

What is it all for? Culling or complementing the workforce?

The workforce discussion is becoming more grounded. The dominant narrative at the summit was not displacement, but redistribution of tasks.

As AI takes on execution, human roles are shifting towards oversight, judgment and decision-making. This is already evident across sectors. In education, outcomes improve when AI is integrated into structured learning processes rather than used as a standalone tool. In enterprise settings, employees are increasingly focused on guiding and validating AI outputs rather than generating them directly. 

This aligns with historical patterns. Productivity gains tend to expand the scope of activity rather than reduce it. The likely outcome is a redistribution of work, with pressure on routine roles but increased demand for higher-skill functions—particularly those combining domain expertise with AI literacy.

The key capability going forward is not technical proficiency alone, but judgment: understanding when AI is adding value, when it is failing, and where human intervention is required.

Conclusion

The AI Summit confirmed that the technology has entered a new phase. Capabilities are expanding rapidly—from robotics to agentic systems to content creation—with clear implications across industries.

However, the gap between what is possible and what is realised remains substantial. Closing that gap will depend less on further technological advances, and more on governance, organisational redesign and disciplined execution.

Additional resources

GSMA Intelligence has various resources enabling clients to better understand AI from a variety of angles:

• AI Inference Calculator
• Telco AI: State of the Market, Q1 2026
• AI for digital industries: what shapes enterprise priorities, expectations and investment plans