Sun Tzu wrote in The Art of War: “Strategy without tactics is the slowest route to victory. Tactics without strategy is the noise before defeat.”
Today, AI is the tactic. Every organisation has access to the same models, the same tools, the same capabilities. What will separate the winners from the noise isn’t the technology they deploy, it’s the strategic systems they build around it. And yet most organisations are still treating agentic AI as the destination, when it’s barely the first step.
Too often AI is deployed like a blunt instrument, call centre automation being the most familiar example, where customers end up more frustrated than before. The technology isn’t the problem. The absence of any intelligent system behind it is.
It’s not just anecdotal, a Harvard Data Science Review study confirms it: the majority of businesses are deploying AI onto workflows designed for humans, not machines. The call centre is just the most visible symptom of a much wider organisational problem.
The answer isn’t to pull the AI out of the call centre, it’s to redesign the system around it. Machines need clear data structures, defined outcomes and unambiguous ownership. Humans need the flexibility to handle exceptions, exercise judgement and course-correct when the unexpected happens. The organisations getting this right are building for both, not assuming one can substitute for the other.
McKinsey’s AI practice, QuantumBlack, puts it plainly: unlocking the full potential of agentic AI requires more than plugging agents into existing workflows, it demands those workflows be reimagined from the ground up, with agents designed in from the start, not bolted on at the end.
Yet for most organisations, that reimagining hasn’t happened, and the numbers make uncomfortable reading. According to the Harvard Data Science Review, 78% of companies claim to be using AI, yet 80% of those report no measurable impact on their bottom line. Adoption is near-universal. Value is not.
Most of that deployment takes the form of chatbots, copilots and AI assistants, useful tools, but ones that sit on top of existing processes rather than transforming them.
The missing ingredient is systemisation, but that word deserves unpacking. The happy path looks like this: a well-designed system with clean data, clear ownership and defined outcomes, where AI handles the repeatable, the predictable and the high-volume without friction. That’s genuinely achievable. Where it gets complicated is at the edges, exceptions, ethical judgements, novel situations, moments where context and consequence matter. That’s where human authority and insight aren’t a workaround, they’re a design requirement.
Knowledge without system is just potential, and potential doesn’t show up on a balance sheet. The good news is that a new generation of agentic AI tools can do far more than answering questions, if the system around them is ready.
Agentic AI represents a genuine shift, from tools that respond to tools that act. Where previous generations of AI waited to be asked, agents can initiate, coordinate and follow through across systems, processes and people. They don’t just surface information, they move work forward. But that capability is only as powerful as the environment it operates in, an agent navigating a fragmented, poorly governed system isn’t intelligent, it’s just fast. And fast chaos is worse than slow chaos.
This is where governance becomes the deciding factor. Not governance in the bureaucratic sense, sign-off chains and compliance checklists, but governance as operational clarity: who owns each process, what decisions can the agent make autonomously, where does human judgement take over, and how is performance measured. Organisations that have answered those questions before deploying agents consistently outperform those that answer them after. The difference isn’t the AI. It’s the architecture of accountability around it.
There is an upside that often goes unsaid: a well-governed agentic system doesn’t just perform, it learns. Every process it runs, every exception it escalates, every outcome it measures becomes data that sharpens the system itself. Done right, agentic AI isn’t a deployment decision you make once – It’s a compounding organisational capability.
Picture an organisation that’s got this right. Their agents run the same process the same way every time, but with enough flexibility built in for local variation. Exceptions are flagged, not ignored. Outcomes are measured at the system level, not the task level. And every cycle feeds back into process improvement. Agentic AI isn’t a tool they use. It’s how they operate.
The proof is in the doing. Consider a company that deploys an AI agent to manage customer onboarding. The agent works, but every team runs its own version of the process. Data is inconsistent, exceptions pile up in inboxes, and no single person owns the end-to-end journey. The agent isn’t the problem. The system is. Now redesign the system first. Define ownership. Map the happy path. Agree where the agent decides and where a human steps in. Standardise the data. Then deploy the same agent into that environment, and the results are transformative: faster cycle times, measurable outcomes, exceptions handled by design rather than by accident. The technology didn’t change. The system did. That’s the point.
Sun Tzu didn’t win by having the best weapons. He won by knowing when to use them, where to deploy them, and what system of command and intelligence sat behind them. The same is true of agentic AI. Anyone who has implemented a major technology change knows this instinctively; success is rarely determined by the technology itself. Ownership, adoption and operational alignment are what determine whether value is realised or simply promised.
The organisations that will define the next decade of business performance are not those that bought the best models or moved the fastest. They’re the ones that stopped asking “what can AI do for us?” and started asking “are we ready for what AI can do?” That means doing the unglamorous work first, mapping processes, establishing governance, defining the boundaries between machine and human authority. It means treating agentic AI not as a technology purchase but as an operating model decision. The battle for AI advantage won’t be won in the model selection. It will be won in the system design. Strategy without tactics is noise. But tactics without strategy? That’s just an expensive call centre!
John Samuel, experienced technologist and business founder of Seventh State
