Slack's multi-cloud AI design gives each feature designated backup models and a routing layer that can respond to degraded service. Its architecture account describes monitoring endpoint errors and latency, diverting requests when thresholds are crossed, and sending a small amount of traffic back to test recovery before restoring normal use.
This makes fallback a maintained operating process. Different providers' errors must be translated into consistent signals, and each feature needs suitable alternatives selected in advance. The design reduces dependence on one endpoint or provider while adding shared monitoring, cost attribution and operational responsibilities across the clouds.
Terms in this piece · Glossary
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
