25. Agentic Workflows in Practice#
This chapter puts agents to work on common research tasks. Each page shows the task, what the agent does, the guardrails that keep it safe on a shared cluster, and how to check the result:
SLURM Jobs and Cluster Workflows: writing, submitting, monitoring, and debugging jobs.
Experiment Management with Agents: logging, sweeps, and checkpoints, building on SLURM Jobs and Cluster Workflows.
Working with Unfamiliar Research Codebases: getting oriented in a new codebase, with KempnerForge as the example.
Refactoring Research Code into Packages: turning scripts into a tested package without changing results.
Literature Review and Data Exploration: finding and checking sources, and exploring datasets.
Multi-Agent Orchestration: chaining and coordinating several agents, and when one agent is the better choice.