Laurent Giraid stands at the forefront of the modern technological revolution, bringing a wealth of expertise in machine learning, natural language processing, and the evolving ethics of artificial intelligence. As enterprises grapple with the shift from experimental prototypes to full-scale production, Laurent provides a critical perspective on how organizations can bridge the gap between technical potential and tangible business value. His insights delve into the structural changes required to move beyond simple chatbots into the realm of “agentic operations,” where AI agents handle complex, repeatable tasks with high degrees of autonomy. In this discussion, we explore the emerging divide between AI leaders and laggards, the critical role of content governance, and the strategic importance of maintaining a flexible, multi-model infrastructure in an increasingly volatile market.
The conversation highlights a dramatic acceleration in AI maturity, noting that the share of organizations identifying as advanced or leading edge has surged from a mere 8% to a staggering 64% in just one year. Key themes include the realization that the primary bottleneck for ROI has shifted from model capability to content access, the transition of governance from a restrictive force to a speed-enabling framework, and the strategic move away from vendor lock-in to favor interoperable, headless agent systems.
Organizations are moving from individual experiments to integrated agentic operations. How do you define the specific characteristics of this transition and what does it look like in practice for a leading-edge company?
The shift we are seeing is fundamentally about moving away from the “sandbox” mentality where AI was a curious toy for individual employees. We have entered an era of systematized, integrated agentic operations where AI agents are actually in production and performing repeatable business functions. This transition is reflected in the data, where the share of organizations calling themselves early-stage or not yet started has collapsed from 53% to just 9% over the past year. In practice, a leading-edge company isn’t just letting people play with prompts; they have built a specific “operating muscle” that involves dedicated teams for deployment and a consistent content layer for those agents to draw from. It’s the difference between a one-off experiment and a reliable, scalable system that delivers measurable business impact within six months of project approval.
What are the primary execution drivers that allow leading-edge companies to achieve significantly higher returns on their AI investments compared to their early-stage peers?
The divide between the tiers is essentially a matter of execution and rigor rather than access to superior technology. Our research shows that 80% of organizations now report a notable return on investment of at least 10%, but the leading-edge companies are hitting much higher benchmarks. Specifically, half of these leading companies reported an ROI above 25%, while only 11% of early-stage companies could claim the same. This happens because leaders treat AI as a core business process, moving past the ad hoc, experimental approach to a structured design. They focus on the “operating muscle,” ensuring that there is formal governance to control these agents and that they are not just isolated tools but integrated components of the enterprise workflow.
With content being identified as the defining bottleneck for 2026, why do so many organizations struggle to connect their agents to the right company-specific data?
There is a significant gap between the recognized need for content and the actual technical execution of connecting that content to AI. While 96% of organizations realize that agents require access to company-specific content to be effective, only 36% have successfully connected those agents to trusted data across multiple use cases. This is largely an issue of trust and security rather than a lack of raw technical capability. Many organizations are finding that their content is fragmented across systems or lack the adequate permissions and access controls, with 21% citing these permission gaps as a major hurdle. Furthermore, 18% of decision-makers describe their content as too unorganized to even make accessible, making the dream of a fully informed agent difficult to realize without a major cleanup effort.
Nearly half of all organizations have already experienced an AI-related data exposure incident. How does a mature governance framework actually help a company move faster rather than slowing them down?
It sounds counterintuitive, but the data clearly shows that 93% of respondents believe better governance is exactly what allows them to move faster. When you have an established or advanced governance framework—a figure that rose from 24% last year to 73% this year—you make the scaling of AI survivable. Governance provides the necessary visibility, yet currently, only 39% of organizations have comprehensive visibility across both sanctioned and unsanctioned AI use. By securing the content layer and ensuring it is highly permissioned, you can run multiple agents across various processes simultaneously, creating a powerful multiplier effect. Instead of viewing a data incident as a total setback, mature organizations use it as a forcing mechanism to refine their permission structures, moving from human-centric workflows to governance built specifically for agents.
There is a growing trend of avoiding lock-in with a single AI vendor. How should enterprises balance the need for high-performance models with the desire for a flexible, multi-model architecture?
The era of “token-maxing,” or simply defaulting to the most expensive and largest model available, is rapidly coming to an end. Organizations are becoming much more pragmatic, with 68% expressing concern about depending on a single AI provider. We see the average number of officially adopted AI tools climbing to 3.3 as companies look for the cheapest model that meets their specific quality bar. This shift toward a flexible architecture is very similar to the multi-cloud movement; it’s about ensuring that every part of the AI stack remains swappable so you don’t have to bet on which individual tool wins the market. By ensuring that 79% of agents can operate headlessly and connect directly via APIs, companies preserve their negotiating power and can pivot as different model families leapfrog each other in performance and cost.
What is your forecast for how the relationship between unstructured content and AI agents will evolve for the average enterprise over the next three years?
Over the next three years, we will see a fundamental reclassification of unstructured data from “dead weight” to a primary competitive advantage, a shift already embraced by 63% of the most mature organizations. Enterprises will move away from retrofitting human permissions onto AI and instead build governance designed for agents from the ground up, tracking every source and permission an agent touches. I expect a widespread adoption of a hybrid token compute budget model, where IT departments manage the core infrastructure while individual business units take ownership of application-level spending. For those who feel they are behind, the good news is that you don’t have to follow a slow maturity curve; by building a multi-model system with a secure content layer from the start, any company can leapfrog into a leadership position. The focus will move from simply having AI to mastering the “content layer” that fuels it, ensuring that every contract, report, and document is ready to be leveraged by a fleet of autonomous agents.
