Lungile Mginqi, digital transformation strategist.
Artificial intelligence (AI) adoption is already widespread. AI transformation is not. The difference is whether leadership is prepared to redesign real work, take people through the change and build the capability to operate AI safely at scale.
Enterprise AI experimentation has become a form of avoidance. AI excitement has done its job. It got executives interested. It is now becoming an excuse to postpone the decisions that actually matter.
Organisations are launching copilots, proofs of concept and training programmes. Those experiments helped executives understand what AI could do. But many enterprises are now repeating the experiment instead of confronting the real work.
Stopping experimentation does not mean stopping learning. It means stopping endless tests of whether AI can impress when the unresolved question is whether the organisation can redesign work around what AI already makes possible.
If AI is expected to deliver significant value, influence decisions or transact through enterprise systems, firing more pilots into the organisation is not a strategy.
The question is whether leadership can redesign the enterprise of 2030 around AI, while building the governance, controls and evidence needed to detect model failure, trace consequential decisions, explain outcomes where required and preserve accountability at scale.
Imagine your enterprise in 2030
By 2030, AI will influence how enterprises price products, allocate capital, serve customers, detect fraud, manage working capital and deploy people.
The World Economic Forum reports that 86% of surveyed employers expect AI and information-processing technologies to transform their businesses by 2030. Stanford’s 2026 AI Index puts organisational adoption at 88%. Yet McKinsey reported in late 2025 that only 7% of respondents had fully scaled AI. Adoption is commonplace, while transformation remains rare.
The enterprise does not need more AI enthusiasm. It needs operating decisions.
The divide will be between those that converted AI into an institutional operating capability and those still managing collections of tools and demonstrations.
What, precisely, are the latter waiting to learn from another pilot?
The next model is not your strategy. Waiting for better AI does not prepare the enterprise.
Leadership should benchmark current capability, test plausible futures and ask what each means for work, skills and operating capability. That learning compounds and cannot be bought overnight.
Pilots preserve the existing enterprise
Most experimentation occurs around the edges of work. Employees use AI to summarise, draft, research and write code. These applications improve individual productivity but do not necessarily transform the enterprise.
The process, decision rights and organisational boundaries remain. So does responsibility: business leaders keep running the business, while the CIO or chief AI officer is expected to find value from AI around it. That is why pilots are comfortable. They let the enterprise experience AI without requiring leadership to redesign the enterprise. AI cannot transform a business that leadership has declared off-limits to redesign.
AI pilots test what the technology can do. Production AI requires the enterprise to decide what people, processes and machines should do differently.
The route to production should begin with a problem large enough to matter: a material revenue opportunity, a significant cost base, a critical customer journey, a regulated decision or a major control process.
Small experiments allow the organisation to admire AI. Significant problems force it to design for AI.
Redesign the work, and take people with you
AI’s real value will come from changing how work moves, how decisions are made and how capacity is created. McKinsey’s 2025 research found that workflow redesign had the strongest relationship with reported bottom-line impact from generative AI, yet only 21% of respondents using it said their organisations had fundamentally redesigned at least some workflows.
People cannot be introduced after the technology has been designed. Future-state design must decide what AI should absorb, what judgement remains human and what higher-value contribution people should move into.
Who directs the AI? Who challenges its conclusions? Who approves consequential actions? How will junior employees develop expertise if AI performs the work through which experience was acquired?
If the work does not change, the enterprise has adopted AI without transforming anything that matters. Automating five minutes inside a five-day process does not transform the process.
Production exposes what pilots conceal
Once AI enters consequential work, the enterprise encounters problems that demonstrations rarely reveal.
What happens when evidence conflicts, a source is stale, a model changes or an agent requests an unauthorised tool? What happens when performance deteriorates after deployment?
AI will not scale simply because the models improve. It will scale when enterprises can trust what the systems are allowed to do, trace what they actually did and stop them when they cross a boundary.
Enterprise-grade AI is not AI that never fails. It is AI whose actions can be observed, whose decisions can be reconstructed, and whose failures can be contained and recovered from.
That requires accountable owners, identities, permissions and operating limits. Material changes must trigger re-evaluation. Executions and human approvals must be traceable. Agents must be stoppable, revocable and capable of rollback.
This is where simulation, continuous evaluation, version control, observability, policy enforcement and audit reconstruction become essential. These are not technical embellishments. They are the control and evidence machinery that allows the enterprise to grant AI more autonomy without surrendering accountability.
Platform selection should follow the future-state operating model, not define it. The platform is a consequence of redesigned work requiring repeatable identity, evaluation, traceability, cost visibility and operational control.
You have already made a platform decision
Enterprises that say they are not ready to make a platform decision create an illusion of neutrality. Architecture deferred is architecture delegated.
AI will enter through Microsoft, AWS, Google, Salesforce, SAP, ServiceNow, Workday, business units and employee-selected tools. Without an explicit architecture, local decisions, existing vendors and accumulated convenience will design the estate instead.
Three broad paths are competing. Teams can build across models and frameworks, gaining freedom but duplicating controls. AI can develop inside existing applications, accelerating adoption while fragmenting intelligence across vendors. Or the enterprise can establish a governed production environment with common identity, evaluation, traceability, cost and operational controls, at the price of concentration.
The credible starting position is to concentrate without surrendering: begin with one preferred production environment, while keeping business logic, policies, evidence and evaluation datasets portable. Start with one. Design for more.
The enterprise does not need more AI enthusiasm. It needs operating decisions. Delegating AI technology is legitimate. Delegating business redesign as a result of AI is not.
The CIO owns the production architecture; the CEO and business leaders must own the redesign of the enterprise around it. The chief risk officer must ensure cyber security, explainability, traceability and auditability are built into the operating model.
The CFO must insist on complete economics, the chief people officer on roles and accountability, and the board on whether institutional capability is being built.
Experimentation showed us that AI can generate, analyse, recommend and increasingly act. It did not show us how the enterprise should organise itself around those capabilities.
Stop AI experimentation. Choose a problem large enough to matter. Redesign the work. Take your people with you. Build the capability to operate AI safely at scale. Otherwise, 2030 will arrive, and your enterprise would have completed experimenting or gambling with its future.
