Reliability
Everything filed under reliability.
- Rate limits are product decisions, and AI agents just became your biggest customersAPI gateways have two very different kinds of limits: ones that protect your systems and ones that define what a customer bought. In 2026, agents that never sleep are testing both.
- A dashboard should answer one question in five secondsMost Grafana dashboards are a wall of every metric someone once found interesting. The good ones are designed like a page in a newspaper: one headline, read top to bottom.
- An alert is a contract with a sleeping personEvery page promises that something is wrong, that it matters now, and that a human can fix it. Break any of those and on-call stops trusting the pager, AI triage or not.
- SLOs for features that answer differently every timeLLM-backed features break the usual assumptions: slow by design, a dependency you do not run, and "success" that a status code cannot see. What to measure instead.
- Your observability bill is a cardinality billMonitoring costs rarely grow with traffic. They grow with labels, log lines nobody reads and traces kept by default. Here is where the money goes and how to get it back.
- Reserved concurrency is the best circuit breaker in AWSMost people meet Lambda concurrency as a limit to raise. Used deliberately, it is a dial that protects your database, isolates noisy functions and works as a kill switch.