October 31, 2026 Your Queued Laravel Agent Starts Over When the Worker Dies The Laravel AI SDK queues your agent as a job with no timeout, persists nothing until the run returns, and keeps its only crash-safe tool handling for approvals. Measured with a worker killed mid-run, then fixed with a tool ledger, a step checkpoint, and a job you own. laravel ai agents laravel-ai-sdk queues
Read article September 18, 2026 Your Laravel Agent Is Not Running the Model You Pinned Model selection in the Laravel AI SDK is a precedence chain with five inputs and no diagnostics. #[Model] kills #[UseCheapestModel], a failover array throws away the model id you wrote at the call site, and the SDK's own test assertion reads a model the request never used. laravel ai agents laravel-ai-sdk cost
Read article September 14, 2026 Your Laravel Agent Re-Sends Every Tool Result It Ever Saw RemembersConversations replays every stored tool result on every turn, up to a 100-row cap the docs never mention. Measured over ten turns on a real agent, with the three levers that cut it and the one that loses your evidence. laravel ai agents laravel-ai-sdk cost
Read article September 12, 2026 Your Laravel AI Agent Costs 5× the Estimate. One Step Is 87% of It. The loop re-sends everything every step, the SDK's total hides which step ate the budget, and caching the history made it worse. Measured on a real Laravel agent. laravel ai agents laravel-ai-sdk cost
Read article July 1, 2026 Observing RAG in Laravel: Your Retrieval Is Getting Worse and Nothing Throws Your Laravel RAG answers 200 OK while retrieval quietly degrades. Here's how to trace retrieval quality, token cost, and faithfulness on live traffic, built from a decorator, a table, and the AI SDK's own middleware. No vendor SDK. laravel ai rag pgvector laravel-ai-sdk observability
Read article