IBM ties intelligent operations to agent safety and business value

IBM is pushing a pretty specific idea here: AI operations should start with the business result, not with the model or the agent. Tyler Lynch, IBM’s outbound chief technology officer for the Automation Portfolio, argued that teams need to step back from narrow build work and think about the capabilities and outcomes they want to deliver for customers and clients. His point was simple and a little uncomfortable for anyone still treating AI as a set of isolated pilots: the work has to connect to the rest of the company.

That matters more as agents take on more responsibility. Lynch described a messy reality where some agents behave like applications, some are autonomous, some are delegated, and some are local co-work agents such as coding tools. Those modes don’t all need the same controls, and IBM’s view is that authorization, logging and telemetry have to reflect that. The company is backing Nvidia’s Open Agent Safety Platform and is contributing enterprise features around agent identity, credentials and hybrid infrastructure.

Lynch also pushed the safety conversation into financial territory. Through discussions with Apptio, the IBM company focused on technology business management, he said agents can emit telemetry for token usage in and out. From there, the question becomes whether policies should evaluate that activity the way identity systems do, and whether an agent’s API key should be revoked if behavior looks wrong. That is a lot less romantic than “agentic AI,” but far more useful when something starts wandering.

Infrastructure context is the other half of the argument. HashiCorp’s Infragraph is now available on a limited basis to eligible Terraform customers, and Lynch said IBM sees it as more than a simple graph of infrastructure. The idea is to turn it into a knowledge graph that can be enriched by FinOps and observability tools. That could help teams catch dependency problems humans miss, especially when a change affects other users or applications that the original team never sees.

IBM’s bottom line is that AI value has to be measured in business terms, not just spend. Lynch said speed of business matters, whether the process is software development, underwriting, or order fulfillment. The useful questions are whether something got to market faster and whether it got done better. That is a much harder dashboard to build than a token bill, which is probably why it’s the right one.