As AI rapidly transforms the financial services industry, a recent City Week conference panel brought together prominent figures from Microsoft, IBM, Quantexa, 3C AGI Partners, and Salesforce to discuss its impact.

Moderated by Tim Hickman of White & Case LLP, the discussion highlighted the immediate opportunities and ongoing challenges of integrating AI, particularly Agentic AI, into enterprise operations.

Rise of Agentic AI

Darren Hardman, CEO, Microsoft UK, emphasized that agentic AI is not a futuristic concept but a present reality. He defined it as a tool that goes beyond human prompts, adapting and reacting autonomously with minimal input, serving as a “brilliant digital assistant.” Hardman outlined four stages of AI transition:

1. consumer efficiency (for example, ChatGPT for personal productivity);
2. task outsourcing (for example, doctors outsourcing patient analysis to AI agents);
3. human-agent collaboration (redefining workflows, with Microsoft seeing 5x productivity gains in internal processes);
4. full business outsourcing (currently limited to a few startups with high agent-to-human ratios).

Zahra Bahrololoumi CBE, CEO, Salesforce UK and Ireland, echoed this sentiment, stating that Agentic AI and digital labor represent profound opportunities. She highlighted examples such as Heathrow Airport’s “Hallie” agent, which processes and resolves 90% of inquiries without human intervention, and Capital One’s use of agents to manage high-volume, high-risk hiring processes, reducing the typical timeline from weeks to days. Salesforce itself handles 85% of its 60 million annual query visits through agents.

Data as the foundation of AI

Vishal Marria, CEO and Founder, Quantexa, underscored the critical role of data. He stressed that regardless of the AI type, its effectiveness fundamentally comes down to trustworthy, explainable, and scalable data. Real-time data, he asserted, is no longer a luxury but a “must” for decision-making. Marria cited HSBC’s deployment of Quantexa’s platform across 56 markets to curate, transform, and make data available at scale for client onboarding and risk management.

He also highlighted the immense potential of unstructured data, noting that a tiny fraction (0.5-1.2%) of this data is currently used in decision-making, despite its rich contextual information.

Navigating challenges and driving adoption

Leon Butler, General Manager, UK and Ireland, IBM, addressed the challenges of AI adoption. He observed a history of “failed pilots” and “science experiments” without tangible outcomes. His primary recommendation was for organizations to pick a specific use case, define a clear goal, and work backward. Other key challenges include:

• IT complexity: Managing information across on-premise, private, and public cloud environments.
• Unstructured data: Only 1% of enterprise unstructured data has found its way into foundation models.
• Governance: The need for clear audit trails for AI-driven decisions.
• Agent sprawl: The challenge of orchestrating multiple agents across various applications (eg Microsoft, Salesforce, Adobe, Oracle). IBM’s Watsonx Orchestrate aims to address this.
• Skills shortages: A significant challenge in the UK, with a reported 7.5 million people needing training, especially in domain-specific AI expertise for areas like procurement, HR, and supply chain.

Global perspectives and future directions

Esther Wong, Founder and CEO, 3C AGI Partners, Hong Kong, offered an Asian perspective, highlighting the differences in investment culture between China/US and Europe. She noted that Chinese and US companies tend to receive faster and larger funding. Wong also delved into the sustainability of AI, particularly the massive energy consumption of data centers. She pointed out that 2025 marks the first year where more AI compute will be used for inferencing (Agentic AI) than training, leading to significant power demands. Her firm specifically invests in technologies that make AI more efficient, focusing on:

• software and algorithm efficiency – developing lighter versions of Transformer models;
• alternative inferencing hardware – exploring non-Nvidia GPUs that are lighter, faster, and more power-efficient;
• sustainable data centers – investigating underwater or space-based data centers, and more energy-efficient designs on Earth, with a notable mention of China’s advancements in nuclear power for this purpose.
• bio-computing: – Wong excitedly shared that 3C AGI Partners had worked with Cortical Labs, which successfully demonstrated a neuron-based computer playing Pong better than silicon chips after only five minutes of training. This “neural computer” aims to combine human intuition with computational power, suggesting a future where data centers integrate both bio-compute and silicon-based compute. She also mentioned the upcoming launch of the world’s first space data center in August by SpaceX.

Trust and responsible AI

The panel universally acknowledged the paramount importance of trust in AI. Bahrololoumi emphasized that as AI becomes integral to daily decision-making (for example, mortgages, healthcare, insurance), it must be grounded in reliable, real-time enterprise data. Salesforce’s Einstein Trust Layer incorporates guardrails and protocols, including transparency support (identifying AI-generated outputs) and toxicity/bias filters.

Bahrololoumi also made a deeply personal point about the need for diverse teams in AI creation to mitigate bias. She recounted an experience with a soap dispenser that failed to recognize her darker skin tone, underscoring that without diverse perspectives in testing and development, AI systems can perpetuate real-world biases with potentially far more severe consequences in areas like mortgage or healthcare decisions. This highlights that responsible AI is not just about technology and data, but also about the people who create and implement it.

AI’s impact on financial services

Hardman concluded by bringing the discussion back to the financial services industry. He cited Barclays’ decision to deploy 100,000 “co-pilots” to create a “super agent” experience for their information workers. This initiative aims to free employees from mundane, repeatable administrative tasks, allowing them to focus on value-adding activities, creative problem-solving, and direct customer engagement. This transformation, he noted, contributes to significant economic growth and enhances the competitiveness of the UK’s financial services industry on a global scale.

The panel’s insights clearly demonstrate that realizing the full potential of AI, particularly Agentic AI, requires a strategic approach that prioritizes clear use cases, robust data foundations, technological innovation, and a deep commitment to responsible and ethical development, driven by diverse perspectives.