Enterprises today are moving beyond isolated AI experiments and asking a fundamental question: what should artificial intelligence actually help the organization achieve? This shift marks the dawn of intelligent operations, where AI agents are not just clever tools but integral parts of a broader business system. The focus has turned to aligning every automated action with customer outcomes, infrastructure realities, and financial accountability. Without this alignment, even the most sophisticated models risk delivering novelty without measurable value. Leaders must therefore start with a clear business intent before diving into technology details, ensuring that every agent serves a purpose that resonates with the organizationโs strategic goals.
Tyler Lynch, outbound Chief Technology Officer for IBMโs Automation Portfolio, emphasizes that the first step for any team is to step back from the role of developer or engineer and think about the capabilities and outcomes they want to deliver for customers and clients. He argues that intelligent operations should rest on multidisciplinary pillars that bring together data, automation, and operational signals into a cohesive framework. This approach encourages crossโfunctional collaboration, breaking down silos that often hinder the scaling of AI initiatives. By grounding projects in desired business results, teams can better prioritize work and measure success in terms that matter to the enterprise as a whole.
Lynch points out that AI agents are not a monolith; they come in various forms that demand different oversight strategies. Some agents operate like traditional applications, while others are fully autonomous, semiโautonomous, or delegated to human supervisors. There are also local โcoโworkerโ agents, such as coding assistants, which interact closely with developers but require distinct controls around authorization, logging, and telemetry. Because no single governance model fits all these modalities, organizations need a flexible operating model that can adapt to the specific behavior and risk profile of each agent type while maintaining overall safety and compliance.
To address the growing need for agent safety, IBM is contributing enterpriseโgrade capabilities to Nvidiaโs Open Agent Safety Platform. This collaboration focuses on establishing strong identities, secure credentials, and consistent controls across hybrid infrastructures. By treating agents as firstโclass citizens in the security stack, companies can enforce policies that govern who or what an agent can access, how it authenticates, and what actions it is permitted to take. This foundational layer of trust is essential as agents begin to handle sensitive data and execute critical workflows across multiโcloud and onโpremises environments.
Building on safety foundations, Lynchโs discussions with the team at Apptioโnow part of IBMโhave highlighted the opportunity to extend agent oversight into the financial domain. Modern agents can now emit telemetry that captures detailed token usage, both inbound and outbound, providing a granular view of computational consumption. This data enables policies similar to those used for identity threat detection and response: if an agentโs token usage spikes beyond acceptable thresholds, the system can automatically revoke its API key, pause its activity, and alert a human reviewer. Such financial guardrails help prevent runaway costs while preserving the agility that AI promises.
Effective intelligent operations also depend on a deep understanding of the infrastructure that agents might influence. Lynch highlights HashiCorpโs Infragraph, which is being extended from a simple graphโbased representation of Terraformโmanaged resources into a richer knowledge graph. This enhanced graph will be consumed by IBMโs FinOps and observability tools, which in turn will enrich it with additional context such as cost allocations, performance metrics, and dependency relationships. By turning raw infrastructure data into a semantic layer, teams gain the ability to reason about the broader impact of any change before it is applied.
One of the most powerful applications of this knowledge graph is dependency impact analysis, a practice Lynch describes as a scientific pursuit aimed at catching errors that humans might overlook. When a team proposes a modification to a specific resource, the graph can reveal other applications, services, or users that rely indirectly on that resource. This capability helps prevent unintended outages or performance degradations that could arise from hidden couplings. As environments grow more complex, automated dependency analysis becomes a critical safeguard for maintaining system stability and reliability.
Measuring the true value of AI investments requires looking beyond simple spend metrics. Lynch argues that speed of businessโhow quickly an organization can bring a product, service, or decision to marketโis a key indicator, though it is often difficult to quantify. Rather than asking only whether an order was fulfilled, leaders should inquire whether it was fulfilled faster and with higher quality. This perspective shifts the evaluation from costโcenter thinking to valueโcreation thinking, encouraging teams to optimize for outcomes such as reduced cycle times, improved customer satisfaction, and increased revenue potential.
Financial operating practices are evolving in tandem with these measurement challenges. Organizations are moving away from static, retrospective cloud bills toward shorter optimization cycles and forwardโlooking forecasts grounded in actual spending patterns. This shift enables FinOps teams, controllers, and chief financial officers to receive timely operational information that can inform budgeting, cost allocation, and investment decisions. By embedding realโtime telemetry from AI agents into financial workflows, enterprises can create a feedback loop where spending is continuously aligned with delivered value.
As companies prepare for events like IBM TechXchange 2026, Lynch advises against attempting to solve every challenge on day one. Instead, organizations should adopt a phased approach: land the foundational capabilitiesโsuch as agent safety controls, infrastructure context, and basic financial telemetryโand then expand gradually to cover more sophisticated use cases like predictive dependency analysis and dynamic cost optimization. This iterative strategy allows teams to learn from early implementations, refine their operating models, and build confidence before scaling across the enterprise.
For leaders seeking to put these ideas into practice, the following steps offer a practical roadmap. First, articulate a clear business question for each AI initiative: what customer outcome or process improvement are you targeting? Second, inventory the types of agents you plan to deploy and map them to appropriate safety, governance, and financial control models. Third, invest in infrastructure knowledge graphs that capture dependencies and enrich them with cost and performance data. Fourth, implement telemetryโdriven policies that can automatically detect anomalous agent behavior and trigger human review. Fifth, establish regular review cycles where FinOps, IT, and business stakeholders jointly assess agent performance against both cost and value metrics. By following this structured approach, enterprises can transform AI from a series of projects into a disciplined engine of intelligent operations that delivers safety, accountability, and tangible business value.