Search Economist Enterprise
By Jim Coleman Director, hybrid cloud and AI technology innovation, Lenovo
Two years ago, enterprise AI strategies centred on experimentation, with success measured through adoption. “Tokenmaxxing”, or maximising AI usage with little concern for cost, efficiency or governance, made sense when organisations were testing AI’s capabilities and limits.
Today, enterprises are under greater pressure to translate AI investment into measurable business outcomes. In its annual State of the CIO report, Foundry, a marketing and research firm, found that accelerating AI-driven innovation and applications had become the top business initiative for 2026, after failing to rank among the top three the previous year.
This reflects what we hear across Lenovo’s AI engagements. Organisations are moving beyond adoption as the primary metric, with disciplined attention to cost, efficiency, governance and value realisation. Executive conversations have shifted from “How do we deploy AI?” to “How do we operationalise AI at scale?”
This shift is being driven by agentic AI.
Unlike traditional co-pilots, agents reason through complex problems, interact with enterprise systems and execute business processes with varying degrees of autonomy. Specialised models, direct connections to enterprise applications and data sources, and more mature governance capabilities are enabling them to plan, reason and execute multi-step workflows across business functions.
As agentic AI becomes more deeply embedded in organisations, competitive advantage will shift to those able not only to deploy advanced systems, but also to trust, govern and continuously optimise them.
Letting results guide infrastructure decisions
Organisations first need to establish whether their existing technology environments can support a growing number of AI workloads and agents without compromising performance, security or governance.
Savvy enterprise leaders recognise that AI optimisation requires a balance of cost, performance and accuracy—and that the right balance depends on the workload. Some applications require maximum accuracy, regardless of token cost. Others prioritise rapid response times or keeping inference costs low at scale. Those business-led choices should shape infrastructure decisions.
Organisations must also determine where each workload is best deployed: in the cloud, on premises, on-device, at the edge or across a combination of environments. The decision should account for long-term inference costs, performance and governance requirements. Enterprise customers rarely standardise on a single deployment model. Instead, they place workloads where they make the most sense for the task at hand.
In this sense, infrastructure strategy and AI strategy have become inseparable.
Beginning with organisational knowledge
AI must understand and reflect the realities of the organisation using it. Large language models have improved general productivity, but generic knowledge alone is insufficient in an enterprise setting. Public models know little about proprietary processes, institutional knowledge or operational history.
This explains why retrieval-augmented generation (RAG) has become so important. RAG connects an AI agent to an organisation's documentation, historical records, policies and business data, helping to ensure that outputs reflect the context in which employees work.
The greatest value comes when that knowledge is brought into existing workflows. Rather than requiring employees to switch systems to interact with AI, well-designed agents can augment existing processes, combining reasoning, relevant information and execution.
For this reason, Lenovo’s AI advisory engagements begin with understanding a customer’s knowledge base, business processes and data before selecting models. In our experience, organisations unlock greater value by connecting AI to proprietary knowledge than by deploying increasingly large foundation models alone.
Making trust an operational capability
As enterprises move from pilot projects to production deployments, they learn that success is not determined by model sophistication alone. It is determined by trust.
Early AI proofs of concept often fail because users lose confidence in the results. Gartner, an IT consultancy, found that at least half of generative AI projects were abandoned after proof of concept in 2025, with poor data quality and inadequate risk controls among the key reasons.
Enterprise AI cannot become mission-critical if employees question its accuracy or consistency. Governance, therefore, can no longer be treated as a final compliance checkpoint. It must be an operational discipline.
Responsible AI requires organisations to evaluate systems before deployment, validate outputs against expected outcomes, monitor for model drift as data changes, and continuously improve performance over time.
As AI systems evolve, governance must evolve through ongoing monitoring, validation and optimisation.
Responsible AI also means investing in people. Teams need the skills to evaluate, monitor and manage increasingly autonomous systems. Organisations that involve their workforce in governance are far more likely to move AI from isolated pilots into trusted enterprise capabilities.
The next phase of enterprise AI will not be defined by the organisations consuming the most tokens. It will be defined by those that operationalise agentic AI they can trust, govern and afford.
Learn how Lenovo Agentic AI can help organisations deploy agentic AI they can trust, govern and afford.
