Subagents Conwidered Harmful
Web page · https://nikvdp.com/notes/subagents-considered-harmful
First seen by Oriel on .
AI summary
The post argues summary-based subagents often lose essential conversational context, while forks and experimental “smart forks” can preserve judgment-critical information with less context bloat.
- In a cited 137k-token Opus 5.5 test, direct work took 67 seconds and cost $0.38; summary-based delegation took 123 seconds and cost $0.57.
- Fresh subagents suit mechanical, well-structured tasks, while work requiring awareness of intent, decisions, corrections, and rejected approaches benefits from a fork.
- Full forks preserve conversation history but also inherit irrelevant logs and side discussions, potentially increasing token costs and reducing model focus.
- The smart-fork experiment uses Cloudflare’s Clef to make parallel keep-or-drop decisions for history pieces, retaining uncertain items and preserving selected text without rewriting.
- Oskar Austegard’s evaluation retained 99 of 105 facts at roughly one-quarter of full-transcript tokens, compared with 102 for the full transcript and 82 for a summary.
#ai agents #context management #jev #claude code
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hacker News: 22, latest observation ; +18 over observed period since
Comments
Comment count reported by the source.
- Hacker News: 8, latest observation ; +8 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hacker News | 50450268 | https://nikvdp.com/notes/subagents-considered-harmful/ Read discussion |