AI agents need more than uptime

Your AI agent can be available, fast and still doing the wrong thing.

That is why monitoring uptime alone is no longer enough.

Splunk and Cisco’s State of Observability 2026, released October 20, highlights the gap.

Among 2,050 ITOps and engineering professionals surveyed across 10 countries:

• 87% say monitoring AI is a priority.
• 56% say they cannot measure AI workload metrics such as response quality.
• 55% say they cannot control AI costs.

These are self-reported findings from research conducted by Oxford Economics in February-March 2026.

They point to an important executive challenge: deploying agents faster than we can evaluate their work.

My takeawayScale AI agents only when you can verify the quality of their work, understand their cost, and intervene when they go off track.

Before scaling an agent, require visibility into:

• Outcomes: Did it complete the task correctly?
• Actions: Which tools, data, and systems did it access?
• Economics: What did each successfully completed task cost?
• Exceptions: When did it retry, fail, or require human intervention?
• Control: Who can investigate, intervene, and stop it?

Human-on-the-loop oversight requires that visibility. A person cannot supervise what the organization cannot observe.

An agent completing more tasks is progress only when those tasks deliver the intended result.

Can your team explain the quality, cost, and business outcome of an agent’s last 100 tasks?