Agentic AI for large organisations: beyond token-maxxing

Key takeaways

• Moving beyond token-maxxing: While some measure AI success by the volume of data processed, true competitive advantage lies in shifting focus from simple productivity gains to measurable business outcomes.
• The shift to Agentic AI: Transitioning from Generative AI—which amplifies individual tasks—to Agentic AI allows organisations to “rewire” end-to-end workflows through autonomous AI agents capable of reasoning and collaboration.
• Trust is the ultimate differentiator: Beyond foundational models, true advantage lies in grounding agents in proprietary data and expert-architected workflows; navigating this complexity requires a trusted partner to bridge the gap between AI innovation and regulatory integrity.
• A framework for scalable transformation: Success in the 2030 landscape requires a structural redesign across six key pillars, integrating automated governance, technology orchestration, and a “value-per-token” economic model.

At the Mistral AI Now Summit, Charles Holive, Chief AI Officer, BNP Paribas CIB, challenged the industry to look past the current obsession with “token-maxxing.” For leaders implementing agentic AI for large organisations, true success is not measured by the volume of tokens consumed, but by real business outcomes: revenue, productivity, and customer impact. Read his key insights below.

Watch the full presentation here

From productivity to business redesign

To navigate the transition to agentic AI, organisations must distinguish between the two current phases of adoption.

The first phase, Generative AI, acts as a “super-power” for individual employees. It uses Large Language Models (LLMs) to perform existing tasks faster and more efficiently. If Generative AI is like giving every employee a high-performance calculator, it amplifies expertise but does not fundamentally change the underlying process.

The second, more disruptive phase is Agentic AI. This is the “rewiring” phase. Unlike standard Generative AI, agentic AI involves autonomous AI agents that can reason, act, and collaborate to execute end-to-end workflows. This is akin to building a digital workforce capable of managing an entire project from ideation to execution, requiring only high-level human supervision.

For the modern enterprise, the challenge is no longer “how to execute,” but “what to build.”

Trust: the only true differentiator

In the agentic era, trust is the primary competitive advantage. While many organisations have access to the same foundational models, the ability to generate unique value depends on the quality and integrity of the information fed into the system. For autonomous AI agents, trust is built by grounding them in high-quality, proprietary data. Raw market feeds, transaction logs, and client records can no longer remain in static silos; they must be transformed into curated “data products.” By building an “intelligence layer” on top of these products, organisations ensure that agents perform accurate queries based on trustworthy, governed information rather than getting lost in fragmented datasets.

This trust extends beyond data to the architecture itself; it is cultivated by the expertise of people who design safe, high-performance agentic systems and workflows. Ultimately, this enables a “value-per-token” (VPT) perspective, where AI investment shifts from a speculative cost centre to a transparent, trust-backed driver of unit economics.

Ultimately, technical reliability must be matched by institutional integrity. For businesses, true success in this transition comes from partnering with trusted advisors who combine advanced AI capabilities with a proven track record of security, regulatory expertise, and predictable outcomes.

This transition is underpinned by the long-standing relationships and trust between clients and their bankers. Charles Holive explains: “For our corporate and institutional clients, the transition to agentic AI is a structural evolution of their operating models. It requires strategic investment and a long-term view. We are here to act as a strategic partner in this journey—helping our clients navigate this complexity.”

The six-pillar framework for agentic transformation

Transitioning from isolated pilots to enterprise-scale agentic workflows requires more than just sophisticated models; it requires a fundamental structural redesign. To move beyond proofs of concept and achieve scalable transformation, Holive advocates for a six-pillar framework—a methodology designed to embed AI responsibly across a global, regulated organisation.

• Business-led strategy: Success requires moving beyond fragmented experiments toward flagship programs—high-value, end-to-end transformations with committed targets for revenue and productivity.
• Technology orchestration: As AI technology becomes a commodity, value shifts to orchestration. Organisations must focus on building robust agent registries and observability layers to ensure solutions can be reused and scaled globally.
• Data as the competitive advantage: Proprietary data is the ultimate differentiator. Data must be treated as a curated “product,” ensuring agents have the intelligence layer and the rigorous lineage required to make autonomous decisions safely.
• Automated governance: In a highly regulated global environment, human-led governance cannot scale alongside thousands of autonomous agents. Compliance must be integrated directly into the software architecture through automated policy engines that provide an auditable trail for every AI-driven decision.
• People and culture: The transition to AI is an opportunity for professional evolution. Companies should focus on providing the training, tools, and support necessary for employees to work alongside AI. By unbundling repetitive tasks, organisations empower their workforce to focus on high-value critical thinking and decision-making.
• FinOps and Value-per-Token (VPT): To manage escalating costs, a pragmatic economic lens is required. Just as an organisation would not assign a senior executive to a task an intern could handle, we must balance high-reasoning models for complex tasks with cost-effective models for routine operations.

Defining the 2030 Landscape

The transition to an agentic future requires a fundamental shift in mindset. As Holive notes: “If what got us here won’t get us there, we must rewire the ecosystem from the ground up.” 

This shift toward Agentic AI will fundamentally change the “nodes and wires” of the financial ecosystem. In the coming years, the most successful organisations will be those that integrate seamlessly into agentic ecosystems—becoming the most efficient nodes that other autonomous systems want to plug into.

Simply being the best version of a legacy organisation in 2025 will not be sufficient for 2030; the objective is to reset today to thrive in a future where interactions are a seamless mix of human-to-human, human-to-agent, and agent-to-agent.