Cursor's SQLite reconstruction experiment illustrates how assigning different models to planning and implementation can change an AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → system's cost. Planners divided the task into smaller jobs; workers implemented them. The agents received documentation while source code, tests, the database binary and internet access were withheld.
Cursor reports that every new configuration eventually passed the held-out query test suite. A run pairing Anthropic's Opus 4.8 planners with Composer 2.5 workers cost $1,339, versus $10,565 for OpenAI's GPT-5.5 in both roles. That supports testing model assignments against completed work on a defined task; passing this suite does not establish complete database equivalence.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
multi-agent — Using several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.
