Reliability

Green Dashboards, Red Users

Every operations person has lived this one. The monitoring says everything is fine, the service desk phones are ringing, and someone senior is asking why the app is “down” when every graph is green.

A Digital Twin of the Workplace

On 9 April my employer, Kyndryl, announced an AI-powered digital twin for the workplace, built on Microsoft Foundry and Azure. It builds a live model of the workplace from signals across devices, applications and locations. Agentic AI then spots emerging problems, suggests fixes, and checks that the fix actually helped. The example that stuck with me was an airline gate agent whose systems are slow just before boarding. Their problem isn’t a server metric. It’s a crowd of passengers and a departure time.

From SLAs to XLAs

This connects to something SREs have argued for years: measure reliability from the user’s point of view. Service level indicators should describe what users experience: page loads, login success, time to complete a task. “Is the server up?” isn’t enough.

Out in the wider industry, that idea is moving from service level agreements (SLAs) towards experience level agreements (XLAs). Commitments are made on outcomes people can feel, not just component uptime.

A few things I’d encourage, whatever tooling you use:

  1. Pick three or four user journeys that matter most, and measure them end to end, from where users actually are, on the devices they actually use.
  2. Break it down by location. One office on a bad network link can be invisible in a global average.
  3. Close the loop. A fix isn’t done until the user’s experience has measurably improved.
  4. Keep humans in charge of the automation. Agentic remediation is powerful, but it needs clear guardrails on what it may change.

(I work for Kyndryl. The views here are my own.)

More Information