Building an adaptive company harness: dynamic agent factories to beat context bloat
hwchase17 · x · 2026-10-23
When every team loads their own skills and MCP servers into a company-wide mega-agent, context bloat is inevitable. Caspar shares the fix in Managed Deep Agents: build the agent from a factory function that runs at the start of each run with all relevant context, dynamically assembling a different model/skills/tools/sandbox/memory setup to suit each team's needs — finance gets billing skills, platform gets an incident MCP server, and nothing else gets loaded.
More from coding & agent
- Artificial Analysis launches Coding Agent Index v2.0 with lower scores across the board — ArtificialAnlys · 2026-10-23
- Coding Agent Index v2.0 scores drop across the board: 20 of 21 configs decline — ArtificialAnlys · 2026-10-23
- SWE-Atlas-QnA-AA benchmark: Muse Wonder Full tops revised repo Q&A grading at 68.8% — ArtificialAnlys · 2026-10-23
- FrontierCode v1.1: Opus 5.5 leads at 56%, mid-pack tightly packed at 43–49% — ArtificialAnlys · 2026-10-23
- Artificial Analysis launches Coding Agent Index v2.0: Claude Opus 5.5 leads at 61.7 — ArtificialAnlys · 2026-10-23
- File-Retrieved Tool Output Makes GPT Agents Substitute Approximations, Costing Accuracy — ucriverside · 2026-10-23