Community guidelines
Be specific and constructive. No vendor spam — promoting your own product belongs in a listing. Anyone can read; posting needs a free account.
Our platform team already runs CI, Kubernetes, templates and observability. Now we've been asked to provide approved models, coding agents, prompt gateways and cost controls for every product team. Is this becoming normal platform engineering work, or are we building a second platform with no extra people??
It is becoming normal, but it is still extra work. We treat model access like any other internal service: identity, quotas, logging, approved endpoints and a paved path. We do not own every prompt or agent. Product teams own behavior and evaluation.
the cost model surprised us. A small group can burn more on long context agent runs than on their dev cluster. Give teams visibility by project and set soft limits early. Blocking after the bill arrives creates bad incentives
The ownership split is our missing piece. Security thinks platform owns all agent risk, product thinks the models are just an API. Who signs off on tools that can write to production??
Same people who sign off on any production automation. Platform provides the identity and policy controls, security reviews the risk, and the service owner approves the actions. Give agents separate identities and require human approval for high impact writes. Do not hide responsibility inside a shared API key.