Run site-extraction work through a low-cost batch agent
A developer sends large extraction jobs to cheaper batch models instead of their main coding agent to keep spending down.
- What they did
- When they need to collect data from a very large set of sites, they create a separate agents-SDK task rather than using their regular coding agent. That task runs on a lower-priced model such as Muse, Deepseek, or GPT-6-luna through a batch API. Routine cloud tasks go to cloud agents, and the dotfiles plus agent context are stored in a GitHub repo so the execution setup can be changed without much friction.
- How it went
- The example mentions a workload that could cover 5,000 websites, but it does not say what the extraction produced. No timings, costs, accuracy figures, or failure details are provided beyond the reason for the setup: saving money.
- Tips
- This approach makes the most sense for repetitive jobs where batch processing is acceptable. Keep a small test run in the workflow so you can confirm the script and estimate the cost before expanding it.
Try it yourself with Claude Code
You are helping me automate a bulk extraction job. From [input file or folder], visit [number] websites and collect [data fields] into [output format]. Use a cheap model through a batch API, and start with a small pilot on 10 sites so I can review the sample output and estimated cost. Stop and ask for my approval before launching the full run or doing anything irreversible, and record any failed sites in a separate log.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Use Claude Code to clean up Obsidian journal notes
A developer used Claude Code in Obsidian to add links to journal entries and turn rough book notes into polished reviews.
Have a coding agent check papers before each experiment
A Claude Code agent with paper search found useful ideas that a plain run missed while tuning a small language model.
Use Claude Code to manage a markdown second brain
Keep a git-backed markdown vault organized by asking Claude Code to answer questions, find notes, add files, and reshape folders.
Use Reddit threads as pre-code research
They used Claude Code with a Reddit MCP server to turn thread scanning into a fast research step before coding.
Use Claude Code to assemble a research workflow
Kevin Yang used Claude Code to draft, code, and test most of a paper’s analysis while he checked each stage himself.
Use Codex to map Reddit complaints by theme
A developer turned Reddit complaint research into a Codex skill that checks communities, gathers quotes, and sorts issues into evidence-backed groups.