Cisco Galileo buy reflects blurring lines in AI observability
Cisco’s Splunk folds Galileo in with its IT ops products, but AI apps and observability introduce a new layer of management that's up for grabs in enterprises.
Cisco will fill an important part of its AI observability portfolio with its acquisition of Galileo, but many enterprise organizations are grappling with who should primarily use such a tool.
Galileo, founded in 2021, markets products spanning AI management categories, including model and output evaluation, production guardrails, and monitoring and observability. Terms of the deal with Cisco were not disclosed; the acquisition is expected to close in the fiscal fourth quarter for Cisco, between April 25 and July 25. Until then, the companies will continue to operate independently. Once the deal is finalized, Galileo will be integrated with the Cisco-Splunk Observability portfolio, according to a company blog post.
"Teams need visibility across the AI stack beyond signals like latency and errors," according to the post. "Observability must evaluate issues like hallucinations and bias, security metrics to detect [and] mitigate business risks, and track cost and usage metrics to ensure clear ROI."
Lacking visibility into AI agent behavior can have ugly results, according to one analyst.
Torsten Volk
"I recently spoke with a data analyst who discovered that their model had used completely fabricated numbers to fill in gaps in financial reports that executives then relied on to make investment decisions," said Torsten Volk, an analyst at Omdia, a division of Informa TechTarget. "Another example: an LLM that kept surfacing a competitor's products in its list of customer recommendations. And then there are the agentic workflows that are endlessly chatty, wasting a ton of money without much -- or anything -- to show for it."
With enterprise AI pilots stalling before reaching production, AI observability has become a critical consideration for enterprises and IT vendors, said Stephen Elliot, an analyst at IDC.
"Tech executives should be asking today of any vendor that's pitching an observability product, 'How do you help me reduce the risk of my AI use cases? How do you empower me with critical data that no one else can get?'" Elliot said.
Cisco's purchase potentially meets a critical need, but how well it integrates the acquisition is something one large enterprise customer will watch closely.
"The real question for practitioners is execution. Can Cisco make Galileo native to the Splunk workflow, or does it become another pane of glass nobody wants to manage?" said Steve Koelpin, principal AI observability engineer at a Fortune 50 company. "The acquisitions that work are the ones that disappear into the platform --the ones that don’t stay bolted on. Every enterprise already has too many dashboards."
AI observability breaks org boundaries
It's hard to govern new tools and capabilities that have just come into existence. Especially since it involves processes that are still new and not fully fleshed out.
Steve Koelpin, Principal AI observability engineer, Fortune 50 company
Glitches in AI agent behavior create problems that span not just reliability and resiliency for IT apps and infrastructure, but also IT security, corporate governance and business management. In fact, Cisco's Outshift incubator has proposed an entirely new set of layers in the OSI networking model, layers 8 and 9, to manage AI agent communication.
"Traditional IT operators do not know where to even begin monitoring for these issues, yet they are the ones mostly responsible for doing so," Volk said.
IT ops has taken charge of AI observability at Koelpin's company so far.
"It falls under the observability team, which is broad and is an umbrella for other teams as well," Koelpin said, requesting that his company not be named because of policies prohibiting him from representing it in the press. But this comes with challenges: "It's hard to govern new tools and capabilities that have just come into existence. Especially since it involves processes that are still new and not fully fleshed out."
This organizational upheaval is reflected in a spate of recent AI observability acquisitions by vendors that fall outside traditional IT management categories, such as ServiceNow's acquisition of Traceloop and Snowflake's acquisition of Observe. Similar fragmentation is emerging widely across enterprise organizations, according to one consultant.
"Observability often sits with platform or DevOps teams because it builds on existing monitoring and telemetry foundations," said Varun Raj, head of private cloud platforms, AI platforms, infrastructure and enterprise workloads for a global consulting firm. "Security gets involved where there's clear risk exposure, and governance comes in at a policy level -- but no single function fully owns it end-to-end."
Ultimately, to make enterprise AI work, these traditionally siloed teams have to come together in new ways, Raj said.
"As systems become more interconnected and decision-driven, observability has to bridge platform operations, security and governance," Raj said. "Forward-thinking organizations are starting to treat it as a shared responsibility, with platform teams providing instrumentation, security defining risk signals, and governance setting the boundaries for acceptable system behavior."
Beth Pariseau, senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism. Have a tip? Email her.
Dig Deeper on Systems automation and orchestration
Torsten Volk
