In its early days, cloud was viewed primarily through the lens of IT. It was a way to host applications, store data and provision infrastructure without having to build and operate everything ourselves. Conversations centred on cost, scalability and the shift from capital expenditure to consumption.
However, by democratising access to computing resources, cloud enabled new industries, business models and customer experiences across SaaS, e-commerce and video streaming. With infrastructure available on demand, the economics of launching and scaling a business changed. A startup could test an idea without a large upfront investment in servers. Software could be delivered as a subscription, continuously improved and distributed globally. Businesses could expand capacity with demand, making it easier to serve markets that would otherwise have been uneconomical. Cloud became a backbone of the digital economy.
The era of intelligent infrastructure
Cloud is now evolving into infrastructure that can produce intelligence on demand, as the idea of an “intelligence factory” becomes a reality. Compute, energy, models and software come together to generate outputs that help interpret information, write code, reason through a problem or plan an action. NVIDIA describes these systems as AI factories, with production economics increasingly measured by token throughput, energy efficiency and cost per token. For language models, tokens are the units used to process inputs and generate outputs.
However, converting this technical capability into enterprise value requires much more than access to models and accelerated compute. The AI use cases delivering the clearest results today are largely horizontal and individual—coding assistants, copilots, content creation and knowledge search. They improve personal productivity, but they do not by themselves transform an end-to-end business process.
Broader transformation begins when AI can work with trusted enterprise data, understand business context and connect securely to applications, APIs and systems of record. That requires governed data foundations, clear semantics, orchestration across models and tools, and security controls that define what AI can access and what actions it can take. As AI moves from assisting a person to executing a workflow, identity, permissions, auditability, human oversight and policy enforcement must be designed into the architecture.
These foundations are also essential to solving AI economics. Cost per token captures only the price of model inference; it does not reflect the cost of preparing and retrieving data, integrating systems, orchestrating agents, applying controls, managing exceptions or involving people. Enterprises therefore need to measure the total cost of a successful business outcome—whether resolving a service request, completing a transaction or accelerating a development cycle—and improve the reliability, speed and unit economics of that outcome over time.
"However, unlocking enterprise value requires connecting this intelligence to business context and the ability to act, while ensuring right security controls and governance."
This is where AI agents come in. An agent needs context from enterprise data, past interactions and memory to understand the task at hand. It also needs tools to interact with business systems and execute actions.
With these capabilities, we begin to move from intelligence to outcomes.
Powering the AI journey within enterprise
Unlocking broader enterprise value from AI requires more than deploying individual tools. Coding assistants, copilots and standalone applications such as voice AI offer useful starting points. Extending these benefits across business processes requires an operating framework, or “harness”, built within, independent of the AI model. For enterprise leaders, three areas deserve particular attention: context, security and governance.
Context: One of the hardest challenges is bringing together information from fragmented systems of record, business tools and SOPs. In addition to the availability of data, an agent also needs to understand how these data interact and relate to one another, and which business rules apply. Enterprises require the right frameworks such as Data platforms and, where appropriate, semantic or ontology layers can make these relationships explicit and the information usable by AI.
Enterprises can begin with the context required for a specific workflow, establishing its quality, ownership and ROI before expanding.
Security: Agents need tools and APIs to act on enterprise information, making control over those tools as important as protecting the data itself. An agent retrieving a customer’s service history requires different permissions from one changing an order or issuing a refund. Each agent should have a defined identity, access limited to its assigned task, and explicit boundaries (incl. sandbox) on the tools it can invoke and information it can share. These controls must be enforced through the underlying applications and APIs, protecting proprietary information and preventing unauthorised actions.
Governance: Enterprises need to define how agents operate across a workflow, where human approval is required and who remains accountable for the outcome. This includes orchestration and reliability: maintaining progress across steps, handling failures without duplicating transactions, and routing exceptions to the right person. An agent may classify a complaint and prepare a response independently, while decisions require approval. Records of actions taken, monitoring of errors and a way to pause or recover a workflow should be built into the operating model. Autonomy can then expand as reliability and business value are demonstrated.
The early mover advantage
For businesses, this evolution opens new possibilities. Coding agents and copilots are already helping teams improve productivity. Beyond this, by connecting intelligence with enterprise context and tools, organisations can increase sales productivity, deliver better customer experiences, accelerate product development and streamline operations – supported by holistic protection of data and IP, regulatory compliance and resilient infrastructure. Given the exponential pace of progress we are witnessing today, early adopters will be well positioned to create new services and business models as capabilities and activities that were once too costly to deliver can now become economically viable at scale, unlocking new sources of growth and competitive advantage.