Tokio Marine Group

Robert Pick

Enterprise AI Strategist

The Enterprise Reality of AI

Few technologies have generated as much excitement, investment, fear, optimism, confusion, exaggeration, and outright nonsense as generative AI. In just four years since its commercial release, AI has become the fastest-moving technology most business leaders have witnessed in their careers. Every board wants it, every vendor claims to haves it, and every analyst predicts salvation or industrial collapse by next Tuesday. Yet for the enterprise, most of AI’s promise still lies ahead.

As a quickly aging, grizzled technology executive, I’ve reached a different conclusion than the evangelists and doomsayers: AI is neither a silver bullet nor an existential threat. It is both more powerful and less magical. At its core, AI is a collection of tools, methods, and capabilities that can create substantial value when applied appropriately, and substantial cost, confusion, and risk when applied indiscriminately. The mistake is treating AI as one thing. It isn’t.

AI can address myriad problems with wildly different risk and value profiles. Thinking of AI as a singular strategy is roughly equivalent to having a “software strategy,” a concept so broad as to be essentially worthless. In plain language, AI resembles a brilliant but distractible teenager. When focused, it can absorb astonishing amounts of information, work tirelessly, and display capabilities impressive enough to elicit parental pride. But when off its game, it fibs, ignores clear instructions, and even gives up.

Navigating the AI Adoption Continuum

Despite these limitations, AI's enterprise potential remains extraordinary. The challenge is understanding where and how to apply it. In a simplified sense, AI adoption exists on a continuum ranging from basic personal productivity to scaled transformational change. At one end, AI summarizes documents, drafts content, conducts research, and returns useful time to knowledge workers. Further along, it augments professionals and becomes embedded within workflows. At the far transformative end, which remains rarely achieved, organizations redesign models and workflows with AI at the center, achieving genuine business model or operational transformation.

This continuum is not necessarily made up of progressive steps but rather different modes of AI use. All modes are valid, and most enterprises should operate in several modes simultaneously. The mistake is overlooking practical value available today while waiting for the continued evolution of AI capabilities, or the necessary funding, to take on a more scaled transformation at the far end.

While massive transformational efforts often operate by their own rules, for more typical AI adoption, there are five key considerations to ponder that are surprisingly easy to skip amid vendor demonstrations, external hype, and internal enthusiasm.

Five Considerations before Applying AI

What problem are we actually trying to solve?

While the habit of indiscriminate slathering AI onto every business problem has calmed somewhat in 2026, AI is still too often the answer before the question is fully asked. Start with the business or technical problem, not the target solution, frontier model, or shiny new agent. If the process is fundamentally broken, AI may simply make it fail faster.

2. What level of AI does this use case actually require, and does it require AI at all?

Not every problem needs an autonomous agent, a frontier model, or a digital army of copilots. Many opportunities can be solved through old-school intelligent automation, workflow redesign, analytics, or even simple training. Use the least-complex, lowest-risk solution capable of reliably achieving the outcome within the realities of the surrounding technology and data estate.

  The silver bullet isn’t AI; it is disciplined thinking about where AI belongs, where it doesn’t, and how to support the people making those decisions.  

3. What is the cost of being wrong?

A misplaced meeting summary and an incorrectly adjudicated insurance claim are not equivalent events. The greater the consequence of failure, the greater the need for explainability, monitoring, safety controls, and governance. The likely blast radius of a mistake should influence both solution design and controls.

4. What does success actually cost?

Token pricing is rarely the right singular metric. Consider the fully loaded cost of a successful result, including development, integration, model consumption, human review, monitoring, support, maintenance, and rework. Cheap AI can become remarkably expensive once operational reality enters the equation.

5. Can we measure this, operate it safely, and govern it effectively?

If we cannot measure outcomes, we cannot determine whether the solution has succeeded. Safety is the ability to observe, monitor, secure, control, and intervene when AI behaves unexpectedly or moves beyond acceptable boundaries. Governance establishes policies, accountability, and controls, then verifies that they remain effective. Can we effectively measure, govern and keep safe our use case?

If a use case survives these five considerations, it is worth pursuing. If not, AI may still be part of the answer someday, or part of a broader solution. It may simply not be the answer today.

Technology Teams Become AI’s First Customers

Ironically, many strong AI success stories are emerging not from business operations but from enterprise technology, although they receive less attention. Development, testing, documentation, modernization, security analysis, and knowledge management have proven fertile ground. Technology teams are becoming AI’s first large-scale customers rather than merely its enablers.

One human consideration is often overlooked: AI removes some work but also creates different work. Someone who once executed tasks may now review outputs, provide context, supervise agents, resolve exceptions, and decide when not to trust the machine. Replacing swivel-chair work with work requiring the skill set of a good secondary school teacher is not necessarily progress, or necessarily welcomed by humans, whom I call “real people.”

Organizations that thrive will understand a simple principle: success is not about deploying the most AI, adopting every new model release, or consuming the most tokens. It comes from applying the right capability to the right problem, under the right controls, at the right cost, while supporting and engaging the people who manage and monitor it all.

AI deserves our attention and ambition. It also deserves the rigor, governance, and healthy skepticism we apply to every consequential technology. The silver bullet isn’t AI; it is disciplined thinking about where AI belongs, where it doesn’t, and how to support the people making those decisions.