NIST AI Risk Management Framework
Primary lifecycle guidance for measuring, governing, and managing AI system risks over time.
← Back to Compute College
Compute College
Observability and GenAI operations
Operate AI systems with metrics, traces, evaluations, feedback, and change management.
Plain-English definition
GenAI operations is the ongoing practice of releasing, monitoring, evaluating, and improving AI systems as models, prompts, data, traffic, and providers change. It combines ordinary service observability with output quality, safety, token, and human-review evidence.
Memory trick: AI operations watches behavior, not only servers.
Why it matters
A stable software deployment can still change behavior when a model or data source changes. Operations must detect quality drift and economic drift, then connect each change to an owner and rollback or remediation path.
- The design makes a hidden tradeoff measurable before it reaches users.
- A clear boundary reduces retries, unsafe actions, or unnecessary capacity.
- The operating choice should be tested against the workload rather than assumed from a demo.
Simple example
A weekly review compares accepted-task rate, refusal and escalation rate, p95 latency, token volume, provider errors, retrieval freshness, and cost per successful task. Changes are linked to model, prompt, context, traffic, and infrastructure versions.
- The example names the input, decision, and expected boundary.
- A deterministic or human checkpoint owns the part a model should not guess.
- The result can be measured with quality, latency, safety, and cost metrics.
Example figures are illustrative calculations, not current quoted market prices.
Current example
Primary reference
This source explains the method or risk boundary. It does not replace testing the actual model, data, provider, and workload conditions in production.
Source discipline: examples on this page are illustrative; measure the real workload before making a release or capacity decision.
Common mistake
Uptime and latency do not show whether an AI service is actually working well. It can be available and fast while producing unsupported answers, excess retries, or economically unacceptable results.
Practical takeaway
What you can do with this
Create a weekly AI operations scorecard with quality, safety, reliability, workload, and cost metrics. Link anomalies to the exact version and change that may explain them.
- Start with a narrow workflow and explicit failure boundary.
- Instrument the path before optimizing it.
- Review the design whenever model, tool, traffic, or policy changes.
Decision check: can the team tell whether a change improved the user outcome without hiding a cost or safety regression?
Compute College learning path
AI Engineering
Step 47 of 48: Observability and GenAI operations