IBM Ties Intelligent Operations to Agent Safety and Business Value

Lynch said no single approach fits every working environment. Some agents are deployed like applications, some are autonomous, some are not and some are delegated, he said. There are also local agents, which he described as co-work agents, including coding agents. Those modes differ for authorization, logging and telemetry, and "no one's nailed it yet," he said.

IBM supports Nvidia Corp.'s Open Agent Safety Platform, contributing enterprise capabilities spanning agent identity, credentials and hybrid infrastructure, according to the interview. Lynch said discussions with the team at Apptio Inc., an IBM company, have raised questions about extending agent safety into financial oversight. He said telemetry can now be emitted directly from agents for token usage in and out, and policies could evaluate that data as they do for identity threat detection and response. An agent's application programming interface key could be revoked to stop it, he added: "Put it in the corner until a human can go look at it."

Acting on operational signals also requires understanding the environment an agent might change. Lynch said HashiCorp Inc.'s Infragraph is now available to eligible Terraform customers on a limited basis. Infragraph is a graph-based representation of infrastructure that is being extended toward a knowledge graph, he said. It will be used by FinOps tools and observability tools, which will also enrich the graph. Shared infrastructure context could help teams assess changes beyond the resources a particular configuration directly manages, because other applications or users may rely on those resources without the team seeing those dependencies. Lynch said dependency impact analysis interests him as a scientific area, including whether a change would adversely affect anyone else.

For intelligent operations, spending alone cannot show whether AI delivers value, Lynch said. Enterprises need to connect that investment to the outcomes of the processes they are changing. "For me, speed of business matters, and that's really hard to quantify sometimes," he said. He described asking whether something can get to market faster, which might involve not only software development but an underwriting process. The question is not only whether an order was fulfilled but whether it was done faster and better, he said.

The financial operating practices supporting those decisions also develop in stages, moving from retrospective cloud bills toward shorter optimization cycles and forecasting based on actual spending, according to Lynch. Those practices can extend operational information to FinOps teams, controllers and chief financial officers. For enterprises preparing for IBM TechXchange, Lynch emphasized thinking beyond individual roles and focusing on customer outcomes. "How do we inform them all? You can't do that all on day one," he said. "We land, and then we expand."