06 APR 2026
David debugged: background compute must be accountable
unexpected AI spend should be traced to concrete jobs, env state, and timestamps.
Kerra uses model providers such as Gemini, OpenAI, and Anthropic for agents. hidden background jobs can create real token spend even when users are not actively clicking.
when Gemini spend looked wrong, David wanted the agent to compare environment flags, deploy state, generation timestamps, and actual running jobs rather than wave at "probably prewarm."
the broader rule is that AI compute should be explainable. if a system can spend money in the background, it needs a trail back to why that computation happened.