While the hype around AI persists, 2026 is shaping up as the year agentic AI will create scaled transformation in financial services. A clear gap is emerging between the market leaders, the chasing pack and the laggards. Visionaries now anticipate the rise of the “10× bank”, where a single individual leads a team of AI co‑workers to deliver exponentially greater output. In this model, growth is no longer constrained by headcount; instead, success depends on an organization’s ability to reinvent work and shape a ‘human‑and‑agent’ workforce with almost limitless capacity and capability. Crucially, people must remain in charge of the change, guiding how these new AI collaborators are deployed and governed.

This series delves in depth into these changes for banking, wealth and asset management, capital markets and insurance — and will cover the following:

1. 1. The competitive advantage of scaling AI: How AI and agentic architectures are reshaping financial services.
   2. Taking a human-centered approach to AI: How AI and agentic AI in particular help us reinvent work and reshape the workforce in financial services.
   3. Leading the change for value: Strategies for driving successful AI transformations, as true business change that is scaled, responsible, rapidly evolving and distributed, and human centered.
   4. Workforce change and the role of HR: How agentic AI is reshaping the workforce and changing the role of the HR function.
   5. Leadership, culture and operating model change: How AI is starting to change how we lead and our ways of working, as it becomes a fundamental capability across our industry.

Moving beyond the hype, scaling for value

In late November 2024, Lord Holmes, David Parker and I hosted an AI symposium for financial services leaders in London. It was evident that AI, particularly generative AI (GenAI), was a change happening now and people were getting beyond ‘POC land’.

Many firms attending were already investing hard in GenAI, using it at scale in production to address real business challenges, often focusing on unstructured data within critical processes like underwriting. All of them were attempting to do it the right way, with responsible AI guardrails and deep care for their employees and customers.

At Sibos in Frankfurt last October, not even a year later and the momentum had exploded. You couldn’t move without encountering organizations using agentic AI to transform the industry. I hosted a panel where we discussed AI as a true co‑worker and teammate—an idea that has rapidly shifted from theory to practice.

We’re driving AI-enabled reinvention in over 2000 engagements, many of these using agentic AI. Our recent analysis of these genAI projects shows that roughly one-third of financial services firms have scaled AI for core processes — and those that have are already seeing outsized returns and accelerating investment. We are beyond the hype and scaling for value — and the gap between leaders and laggards is widening fast.  

Real-world example: software engineering and legacy replacement

One of the standout moments at the symposium was the presentation of an agentic architecture case study, at a time when the idea was still gathering initial interest. We were working with this bank using agentic AI to help our engineering teams accelerate a major legacy-system migration.

Legacy replacement is a critical business issue for resilience, cost and agility in banking. The bank faced poor legacy code quality, with 40% of the code never reviewed. Given this complexity, the initial plan required 250 developers for three years, with bottlenecks already emerging around senior developers, legacy code and domain knowledge.

To address this, we deployed AI agents to work alongside our software engineers:

1. 1. Software development agents: Take user input and requirements, referencing the legacy platform to write new code in a modern language.
   2. Critique and testing agents: Review the code, test and debug it, provide feedback to the development agents (visible to the engineer) and ensure it meets company standards.
   3. Improvement agents: The agents plan and iterate until an acceptable code quality is achieved, with human engineers able to direct and prompt further cycles.

This agentic architecture was designed in collaboration with the software engineers and integrated into their developer workbench, enhancing productivity and elevating their roles. The results were remarkable:

• • Increased speed and efficiency: Development became 30% more efficient — saving approximately £15m — but also accelerating delivery and removing bottlenecks.
  • Improved quality: Code review frequency increased, and standards for performance, security, maintainability, and reusability were raised.
  • Wider benefits: Documentation improved by 40%, metadata coverage by 35%, test generation by 40%, and rework reduced by 25%.

Needless to say, this example was a ‘wow’ moment for the attendees at the symposium. Since then, the progress made last year with agentic AI has become even more impressive and widespread.

The impact in technology delivery isn’t just in engineering, for instance we’ve made data pipelines 98% more efficient for a large Asian bank and for a large European bank we’ve made the service desk 20% more efficient. 

What is agentic architecture?

AI agents are programs that handle tasks and workflows to achieve specific goals. Humans set the goals, but the agents operate more independently, adapting their strategies as needed to achieve the goal. They take inputs, reason, decide on tasks to perform, interact with other agents and tools, review their outcomes and determine the next steps required.

The agents ‘know’ a domain context in which they have been trained within an organization (e.g. on specific training data and documents). They have long-term memory and can learn from past interactions to optimize decision-making.

AI agents are specialized, trained for specific roles and goal oriented. The benefit of this specialization is improved performance, because different agents can be mixed together to form a total greater than a sum of its parts (akin to a multi-disciplinary team).

In agentic architecture, orchestrator or supervisory agents manage the process, assigning tasks to specialized utility agents.

Hang on, how is this different to traditional AI and generative AI?

‘Traditional’ or ‘classical’ AI typically uses machine learning and skilled data scientists. It’s been used for many years in FS for complex models. It is incredibly good at prediction and identification of the next best action (i.e. patterns, forecasts, models, simulations, optimization, recommendations) within defined parameters and inputs.

GenAI creates content based on a prompt into a large language model to produce an output — most of us are used to ChatGPT, CoPilot, Claude or other models now. GenAI models are increasingly powerful and even able to produce multi-model outputs such as film or games. A skilled user can prompt in different ways or in a chain to refine these outputs. However, these models in their vanilla form cannot handle dynamic tasks and execute multi-step plans — they can produce, but it cannot work towards a more complex goal.

Whereas in agentic AI, multiple AI agents collaboratively pursue and achieve more complex, multi-faceted goals with less human intervention at each step.

This is the move from passive content generation (GenAI) to task-specific execution (AI Agents) to more autonomous multi-agent orchestration (Agentic AI) (see Sapkota 2025 for more on this).

All preceding forms of AI (e.g. machine learning, deep learning, GenAI) remain relevant, but it is agentic AI’s capacity to reinvent work by ‘doing things together’ that is significantly greater.

While the software delivery example is interesting and highly relevant to FS, let’s explore some further examples to bring this to life.

Real world examples — Know Your Customer

We’re often asked whether agentic AI can be applied to risk and compliance? The answer is yes! We’re helping a number of banks transform ‘Know Your Customer’ (KYC) using AI.

In the past, KYC has been a slow and costly manual process with legacy systems, with challenges around detection rates and false positives, and knock on impacts to client onboarding and periodic reviews.

Agentic AI allows the KYC process to be reimagined, no longer bound by sequential processing. KYC analysts are no longer focus on highly manual activities and can spend their time on high value judgement-based investigation work instead.

Here are three examples of how the transformation of KYC using AI has evolved over the last couple of years, moving from self-contained GenAI use cases, through LLM workflows, to an agentic end-to-end process.

At a European bank, one of the earliest uses was to classify documents, ingest and extract KYC data points, validate and remediate missing data, then present the data in a consistent format for the KYC agent to validate. For complex correspondent banking cases this reduced ingestion time by 99% and reduced costs by 94%, while raising quality, of these tasks.

At one of the global banks starting in wealth, we started with language translation (saving 90,000+ hours pa), KYC agent guidance (reducing case rejection rate), document classification (95% initial accuracy) and party identification (reducing time taken by 50%) and case summarization etc. These were more expansive uses across workflows.

Most recently, at another global bank, we’re applying agents across the end-to-end lifecycle starting with source of wealth (a major business challenge) where the agents can extract relevant information from documents, identify where there are missing documents or data, generate a source of wealth narrative and review that narrative for accuracy and completeness. The human KYC analyst is still in the loop and has ultimate control over the case processing, but can uplift productivity, accuracy and client experience using these agents. Using agentic AI we can support more of the full value chain.

Real world example — Claims

For a European insurer we are developing agentic AI for claims in property, casualty and motor insurance claims. Claims have a significant economic and customer impact if the wrong decisions are made. For many insurers claims are a labor-intensive process, with lots of time used handling unstructured data and documents.

We developed AI agents to sit on top of the claims platform and orchestrate out to other systems, supporting the human claims handler. The AI agents support tasks such as information gathering, data quality, policy and coverage checks, policy detail extraction, claim data summarization (including the first notification of loss, customer and expert inputs, customer correspondence) and even some support to litigation, fraud and reserve management.

Through all these tasks the agents support the claims handler to be in charge as the ‘human in the lead’ — for instance, they can look back into source data easily. We have been able to free up 20% of their capacity, focusing their time on better decision-making (improving claims accuracy by 1%) and supporting negotiation in the most complex and valuable claims.

We’ve seen similar work from a US-based P&C insurer through to complex commercial claims operations. 

The ‘so what’ for work and workforce

Imagine agentic AI as your co-worker, where you have team members dedicated to you who specialize in their support to you and can even bring perspectives and strengths complementary to your own.

As a business leader, imagine having a workforce, not of 100,000 employees, but of a million human and agent workers with significantly greater capacity and capability, doing work in a radically different way.

Getting started with agentic architecture

To realize the value of agentic architecture, the right data foundations, responsible AI frameworks, infrastructure and skills are essential. We have helped many banking and insurance clients implement these across their businesses, including working with executive teams and boards. The most critical step is identifying the right business opportunities and value streams, involving a workforce who understands these areas best and taking a human-centered approach. To accelerate this, we have assets to analyze and reinvent key banking and insurance value chains, processes and workforces. This includes ‘ready to go’ agents developed on Accenture Refinery, based on NVIDIA and partnerships with the leading AI ecosystem players.

An example of AI as a business change at pace and scale

JPMorganChase provides a particularly interesting case example of rotating to AI quickly and at scale. They have democratized self-service access for 200,000 employees to their LLM Suite in less than a year, half of whom use it 3+ times a day.

This has led to widespread ideation, experimentation and application (within a safe space and at favorable marginalization of cost). They can then harvest the good ideas, investing and scaling them with a ‘venture capital’ mindset.

They now have specialized AI and agentic AI across different business units. This distributed approach to change is not a laissez faire free-for-all.  All models are registered and controls in place to support safe scaling.  However, there are differentiated paths for higher risk uses (separate risk committee, full model review) vs. low risk uses (sponsor reviewed).

Crucially, JPMorganChase have invested in adoption, creating space for high agency early adopters and helping followers and mass adopters get onboard too. They encouraged work disruption: “Let AI eat your job; we have lots of other jobs here for you to do. Your job won’t be taken by AI; it will be taken by a person mastering the use of AI” (Mary Erdoes, CEO of JPMorganChase’s Asset & Wealth Management (AWM) division).

Their high growth, high-talent environment has allowed them to create a meritocratic, but human-centered approach to engaging their workforce in the change. This was about moving rapidly, but also having a plan for workforce change: “The more I know about it, the more I can plan for it, let attrition be my friend, and where necessary, redeploy, retrain, etc.” (Jamie Dimon, CEO, JPMorganChase). 

Key takeaways

At the end of each of these blogs I’ll include some key takeaway points and questions — feel free to use these within your team or let me know if you’d like to chat through them.

1. 1. Leadership and value: Are you leading or lagging on scaling for value?
   2. Business opportunities: How is AI reinventing your business?
   3. Agentic AI: What work can AI do across your biggest value chains and processes? 

Looking ahead

In my next blog, we will explore how leading financial services firms are approaching AI in a human-centered way and how we can use agentic architecture to accelerate the reinvention of work.

For a deeper view of how agentic AI is changing work across financial services, read our Top Banking Trends for 2026 report.