When one agent should become several specialists
Keep one agent on the whole task until the work breaks into specialties that need separate context and instructions. Mikiko Chandrasekhar put a practical size on that split: small agents equipped for one to three tasks, each judged on criteria for its own skill set. After the split, log every handoff so you can see who was called, what crossed between them, and what came back. Mikiko…
When a single agent is asked to perform multiple tasks, it is better to split it into several specialized agents. Mikiko Chandrasekhar, Staff Developer Advocate at MongoDB, suggests keeping one agent on a task until it becomes too complex for a single agent to handle. Each specialized agent should be equipped for a small slice of the task, with its own criteria for performance.
Once the task is split, log every handoff between agents to track who was called, what crossed between them, and what came back. This allows for easier debugging and inspection of the system's components. Mikiko treats agents as software products, similar to how software engineering and MLOps practices are handled. She recommends writing different checks for each skill of an agent and comparing them against the job's criteria.
For coding or copilot-style agents, where the output is code, she suggests measuring signals such as accuracy, runtime efficiency, and acceptance rates. Vibe checks are acceptable up to a point, but quantifiable metrics are necessary for improvement. To make agent-to-agent handoffs observable, log the payload that crosses between agents, along with the final answer.
This creates a view of the agentic graph, showing how agents communicate, data passes between them, and tools are called.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.