Splunk is extending observability to AI token usage to help enterprises answer a question their provider’s bill alone can’t resolve: What was the spending for, and was it worthwhile? Tokenomics was a focus at .conf26 in Denver this year. Cory Minton, Splunk’s Global Field CTO, argues that leaders need visibility into consumption across users and projects to connect AI costs to the work their teams are doing.
In a conversation with Frontier Enterprise in Denver, Minton explained how Splunk is approaching that visibility and how Cisco is using the technology to understand and govern its own AI usage.
What was the development process for Splunk’s tokenomics solution like?
Tokenomics as a practice is an evolution of what we did a decade ago when we started adopting cloud computing. Everybody was saying, “Cloud’s great. It’s going to make IT so much more agile. We can deploy things faster.” Then everybody got the bill and said, “This is more expensive than my data centres!” Now we’re facing a similar challenge with AI: Frontier companies are deploying the technology, but it can be really expensive. We have to understand those costs because IT and security teams will ultimately be on the hook for how much the company spends on AI tokens.
At Splunk, we’d been building the ability to instrument frontier AI capabilities, whether they’re deployed and consumed through an API, used in an IDE (integrated development environment) on your desktop, as most of us do, or deployed in a data centre. How do I understand what it takes to serve my users’ token requests, whether the costs relate to the underlying infrastructure or purely to API usage? Where is the token spend happening? Is it in the reasoning? Is it in the context?
Splunk has always been a data company focused on instrumenting digital systems to understand what’s happening inside them. AI is no different. Tokenomics involves instrumenting the entire stack needed to serve a token request to understand how much it costs and why. Where is my spend happening? Who are the users? From there, we can ask why they spent it. Is the spending connected to a business outcome or a meaningful project that I want my team to work on? The aim is to understand what’s happening in the underlying systems and provide business context so leaders can understand and guide their teams’ use of the technology cost-effectively.
We’ve taken several steps, including releasing an open-source project called Token Meter. It runs on the client side, on your Mac or other desktop device, and provides instrumentation to help you understand your personal token utilisation and connect it to projects. For example, it provides context such as, “This prompt appeared to be tied to this project. When it ran, it used 30% of your tokens for context, 35% for reasoning, and 40% for rendering.” This visibility into costs helps developers and people using AI for productivity make informed decisions about their own usage.
The organisation needs telemetry on every person using those AI tools, which is where the AI observability tokenomics offering came from. We use instrumentation that Galileo, which we acquired, has been developing for years. We integrated that instrumentation into Splunk’s observability capabilities and now aggregate token utilisation data in our platform. That lets us examine not just the size of the bill, which you can get from your provider, but why those costs were incurred. What was the usage connected to? What business value came from that user’s token consumption, and can I have a conversation with them about their usage? Can I put guardrails in place to guide them towards a different use case?
AI enables our users, including analysts and site reliability engineers, to do things they couldn’t previously do, but ultimately, it’s still a business conversation. We have to understand token utilisation in meaningful terms and give leaders visibility into how their teams are using AI.
Cisco is using this tokenomics solution. How has it been so far?

Our IT organisation supports thousands of users globally. We can use tools from Anthropic, as well as Codex and Cursor, but Cisco has also deployed its own hosted agentic system called MyAgent. Its AI tools are connected to our data systems so users can work with agents in their respective roles.
Like any organisation, as we started rolling out these tools, we realised we were spending a lot of money on them. Our IT organisation came to Splunk for help, and our VP of IT said, “Okay, we’ve got this Galileo technology. Let’s deploy it in our environment and start to understand usage and guide teams.”
We now understand usage across tools that make API calls and third-party LLMs. We can also see whether users are sending data they shouldn’t and stop them from sending confidential data to third-party tools that shouldn’t receive it. On our internal systems, we understand token costs and infrastructure utilisation, allowing IT teams to provide governance and advise leaders.
As a leader, I can see my team’s AI usage and ask questions such as, “Walk me through why you had this long-running task that consumed US$1,300 in tokens. Was that worthwhile? What were you doing?” We can then discuss that usage as leaders. Everybody at Cisco has visibility into our token spend, helping us orient ourselves as managers and understand where the tokens are being spent.
Can you provide some background on how Splunk is working with Anthropic?
Anthropic is both a customer and a partner of ours. We use Anthropic’s tools internally at Splunk, and we’re working together to make it easier for Anthropic’s customers to connect their environments to Splunk and use it as a system of record and a source of operational intelligence. We’re doing this via an MCP (Model Context Protocol) connection.
We’ve also been collaborating on skills to contribute to the Anthropic ecosystem, with the aim of making software development more secure and observability capabilities more consistent.
We recognise that many organisations are going to use Claude as the interface to other data capabilities in their environments. Our work together focuses on making it easier for Anthropic’s customers to use Splunk and get operational value from it through that interface.
How is Splunk positioning itself in the cybersecurity landscape following Mythos?
Cisco was among the early adopters of Mythos and Daybreak. While I can’t discuss much of that work, we looked at using those capabilities to protect our customers. Following Mythos, we expect attackers to become more capable and their attacks more advanced and autonomous. How, then, do we defend at machine speed?
We’ve been building tools to instrument those environments and use the same kinds of tools being used to develop attacks for defence. How do we eliminate those vulnerabilities as rapidly as possible? We’ve been working with our Cisco counterparts to gain visibility into attacks as they happen and put enforcement measures and guardrails in place to protect potentially affected environments as rapidly as possible.










