Run async code-research benchmarks with a coding agent
- Source date:
- Last verified:
- Method:
- How we verify
Evidence: Screenshots or logsThe source shows screenshots, a transcript or the agent's output.
Outcome
An async coding agent cloned several Python Markdown libraries, benchmarked them against cmarkgfm, and wrote a report with charts, with almost no human time spent.
Agent
Claude Code for web, OpenAI Codex
Prompt Original prompt
Exact prompt as published in the source.
Create a performance benchmark and feature comparison report on PyPI cmarkgfm compared to other popular Python markdown libraries—check all of them out from github and read the source to get an idea for features, then design and run a benchmark including generating some charts, then create a report in a new python-markdown-comparison folder (do not create a _summary.md file or edit anywhere outside of that folder). Make sure the performance chart images are directly displayed in the README.md in the folder.
Source
https://simonwillison.net/2025/Nov/6/async-code-research/
By Simon Willison · Published Nov 6, 2025 · Added Oct 15, 2026 · Last verified Oct 15, 2026
Related cases
- Compare top social media tools in a spreadsheet with ChatGPT agent and ClaudeRecreated promptEvidence: Screenshots or logsBeginner-friendly
Allie K. Miller asked both ChatGPT Agent Mode and Claude's Excel feature to compare top social media tools, and Claude finished in 3 minutes vs 8 with denser decision support, though some prices mixed annual vs monthly.
ChatGPT agent, Claude (Excel) · Research · Source date:
- Extend a family tree two generations with Claude Code and browser agentsRecreated promptEvidence: Shown on video
Working from the family's archive scans, Claude Code with memory and the Claude browser plugin cross-checked church books, ship manifests and naturalization records, built a 300-page family wiki, and found two more generations in a day; the family is now checking it for errors.
Poster says:300-page family wiki
Claude Code, Claude in Chrome, Perplexity Comet · Research · Source date:
- Find hard-to-locate company tax IDs and addresses with GPT-5 ProEvidence: Described by poster
Working for over 45 minutes, GPT-5 Pro found correct FEINs for 4 of 6 subsidiaries on the Florida Division of Corporations site and one address via a FINRA BrokerCheck report; the poster double-checked the FEINs.
GPT-5 Pro · Research · Source date: