Skip to content
anythingai.space
Deployment & Deliveryintermediate11 min · 10 steps

Feature Flags

Choose code paths, prompt versions and models per request with runtime rules — to release, target, experiment and switch things off without redeploying.

users on v420%
only leverwrite callsDeployminutes each, riskyrag-apibehaviour baked inTicketing toolslow · timeouts
dataerrorrequest
metrics are illustrative
1 / 10
ProblemStep 1 of 10

Every change of behaviour needs a deploy

To try prompt v42 on a few users, the team must deploy a build that contains it — and then deploy again to widen it, and again to roll it back. When the ticketing system slows down, the only way to stop the agent's write tools is an emergency redeploy.

Deploying code and releasing behaviour are the same event, so every release carries deploy risk and deploy latency.

Lesson 15 of 20 in Deployment & Delivery