Splunk by Cisco Brings Agent ROI to AWS Marketplace and Real-time Evaluations to Everyone

We are living in the era of the agentic teammate. Today’s AI agents are working alongside us now, helping to automate and streamline key business workflows and customer interactions. However, as we assign AI agents with more ownership in driving mission-critical environments, organizations are navigating the unpredictability that comes with this shift—both in costs and business outcomes.

Each AI prompt, retry, and token an AI agent generates can trigger unexpected charges. In the State of Observability 2026 report, 55% of respondents said their organizations can’t control AI costs, and 60% said surprise billing spikes strain operations more than high baseline costs.

Cost is only half the risk. Agents can hallucinate or give answers that conflict with company policy. And many companies are rushing to deploy AI faster than they can monitor it, leaving them unable to connect their AI programs to tangible business results. Eighty-seven percent of respondents confirm monitoring AI is a priority for their organization; but 54% still don’t have clear business outcomes tied to it.

To effectively track agent ROI and stop costly mistakes before they happen, companies need a new class of behavioral metrics tied to agent decisions and business objectives.

At Cisco Live Melbourne today, Splunk by Cisco is unveiling new Splunk Agent Observability innovations that bring agent cost and quality into one unified view. They pair agent cost with business ROI metrics, helping to grade every agent interaction and let teams build custom guardrails. In addition, Splunk by Cisco is unveiling a new way to use Agent Launchpad to let developers call Splunk agents from Cisco AI Assistant, Codex, and Claude Code.

Unified Real-Time Visibility Into Agent Cost and Output Quality

Splunk Agent Observability’s enhanced Tokenomics capability will provide comprehensive cost governance that ties AI spend to output quality, giving organizations running on Amazon Web Services (AWS) a unified way to monitor each agent by its cost and business results.

Tokenomics pairs the cost of every agent with evaluations that grade the quality and impact of the work, giving organizations a single view of AI ROI. The solution leverages usage data from Amazon Bedrock and Amazon CloudWatch, then combines those technical metrics with customer-defined business KPIs like NPS (Net Promoter Score) or cost-per-claim to produce ROI reports executives can act on.

This extends the token-spend attribution launched at .conf26, which tracks costs across AI agents and coding tools such as Claude Code, Codex, and Cursor

The new Eval-G1 evaluation models deliver 2.5x faster evaluation response times than traditional large-language-model (LLM) judges—with up to 95% accuracy. This lets teams evaluate and guardrail 100% of their traffic without the bill shock.

When an AI agent serves a customer, teams want to verify agent behavior and enforce their own rules. The Eval-G1 evaluation models make this possible with the speed, economics and precision of a small-language-model (SLM). Together with the new evaluation tuner in Splunk Agent Observability, teams can turn their agentic traces and SME expertise into personalized SLMs that evaluate, steer, and block inaccurate agent behavior at machine speed.

Splunk Agent Observability's Eval-G1 evaluation models and evaluation tuner form a new evaluation engine that empowers teams to monitor and control agents without sophisticated data pipelines or data science expertise. This is another example of how Splunk by Cisco is simplifying workflows to make evaluation practical and affordable so customers can trust their AI at scale. The custom evaluation models and the evaluation tuner are available starting this month in Splunk Agent Observability.

Call Splunk Agents From the Tools Developers Already Use

Starting later this month, Splunk Cloud developers can build, deploy and call their Agent Launchpad custom agents from Splunk AI Assistant and frontier agents like Codex and Claude Code through the Splunk MCP server and API, and through Splunk Command Line Interface and web access (in coming months). Agent Launchpad already lets teams build custom agents inside Splunk: they connect tools and knowledge bases, apply skills, choose a model, and set human approval steps. Splunk alerts can trigger those agents to start the investigation, and searches hand them the context they need.

Beyond these innovations, Cisco is democratizing defense for the AI era by providing teams with proactive, governed defense that scales with their AI operations. Through expanded Enterprise Security Essentials and Premier editions, alongside new Cloud Connect support, Splunk by Cisco is lowering the barriers to adopting autonomous security operations.

Meet us in Melbourne or at KubeCon North America

Splunk by Cisco teams are at Cisco Live Melbourne and at KubeCon North America in Salt Lake City this week. Stop by our KubeCon NA Booth #105 or our Cisco Live Melbourne booth for more information on these innovations.