Code Mode Without the Interpreter: Agents That Write Aggregation Pipelines
MongoDB, Thursday, October 8th, 2026
MongoDB shows AI agents writing validated aggregation pipelines instead of running model-generated code in an interpreter.
MongoDB's Daniele Tassone describes a contract-analysis agent that writes MongoDB aggregation pipelines instead of running model-written code in an interpreter.
A server-side handler validates each pipeline, injects a tenancy filter, runs it and returns only the reshaped result, replacing about five sequential findOne calls and hundreds of KB of context with a few hundred bytes.
Even a 60-file cross-document comparison runs as a single pipeline, an approach that works because models know aggregation syntax well and the language is small enough to allow-list.