Eval gate passed, production broke: add a smoke eval to agent deploys
A 96% offline score, a 3-second timeout and 212 double refunds. The 20-case, 3-minute check that stops it at 5% traffic.
Latest articles
Authorise AI agent tool calls from the JWT, not the prompt
The JWT was valid. The model still picked the customer ID. Move identity into the state so a fooled model can't reach the wrong data.
Idempotent Tool Calls: Stop Your AI Agent From Charging Twice
Retries come from four layers and can multiply to 18 runs of one action. Seven rules make a dangerous tool call happen exactly once.
Single agent vs multi-agent: check the context before you split
In my RAG eval, [X] of [N] failures never had the evidence in context. A second agent would have failed the same way at about 4x the tokens.
Rate limit LLM tokens with Redis when the cost arrives after the call
Reserve an estimate before the model call, correct it after, and give every agent to run its own budget.
How to cache an LLM agent with Redis without serving wrong answers
Three caches in front of an agentic RAG system, and five cases where a cache hit must be treated as a miss
Multi-Tenant RAG Leaks Through the Search, Not the Login
One RAG pipeline, many users: four checkpoints that keep every answer inside the asker's own documents. Grounded RAG in Production, Part 2. When one RAG pipeline serves many teams or customers, the lo
Secure a RAG API without minting tokens or storing passwords
An identity provider issues the tokens, the API only verifies them, and neither the model nor an uploaded PDF gets a say in who the user is.
