AI agent sandboxing for enterprise products: requirements, gaps, and where MXC fits
AI agent sandboxing for enterprise products: the requirements we see, where containers and microVMs fall short, and what Microsoft MXC changes.
Artificial Intelligence
Tool calling: when you need Jev and when LLM caching is enough
Prompt caching cuts the cost of resending tools and context. It does not make a tool-call decision typed. When a Jev gate earns its place and when it doesn't.
Discovery to hypercare: a delivery shape for AI programs
What good AI project delivery looks like: phases, exit criteria, and ownership from discovery workshop to hypercare and ops handoff. Built to end pilot-forever.
Approval gates for AI agents that touch money or citizens
When AI agents can move money or change citizen records, drafting is not enough. Approval gates, spend limits, and action whitelists for controlled autonomy. Bluelupin practitioner notes.
LLM benchmarks: what each bench actually measures
Capability evals and systems benches answer different questions. A plain map of LLM benchmarks for eng leads choosing what to measure before they pick a model or agent stack.