Over 100K context, subagents cost more: Alezander9 favors context retrieval
Alezander9 · x · 2026-11-07
Alezander9 concludes that if the first session exceeds 100K tokens of context, the subagent approach becomes costlier, so he recommends the simpler context retrieval route with a higher ceiling instead, suggesting it be added to Hermes. Teknium added his sessions regularly exceed 100K and invited building this into a plugin at NousResearch's hackathon.
Related event: Teknium: Context retrieval beats subagents for sessions over 100K tokens(3 posts)→
More from coding & agent
- Dev gives on-call agent long-term memory with Hindsight — here's what changed — aryan_tyagi_ase · 2026-11-07
- Developer shares personal tier list of AI coding agents and harnesses — _jaydeepkarale · 2026-11-07
- Rich author Will McGugan: AI-era bug reports are like full-body scans — hugobowne · 2026-11-07
- Dev argues agent skills are mostly slop, prefers a single master agent instead — teodorio · 2026-11-07
- Podcaster edits a 90-minute episode with Claude Opus 5.5 in about an hour — audrow · 2026-11-07
- 10 open-source repos to build an AI content factory, from OpenCreator to FLUX.2 — Aiden_Tech_Ai · 2026-11-07