'Subagents vs Agent Skills' compares loading skill packages as instructions into the main context versus launching them as subagents with dedicated context windows. Subagent execution wins when packages have clear I/O contracts and procedural instructions; skill-loading grows brittle as task horizon increases (context accumulation); the cost is coordination tokens.
Subagents beat context-loaded skills as task horizons grow
2026-09-11Frontier watchTRIAL
'Subagents vs Agent Skills' compares loading skill packages as instructions into the main context versus launching them as subagents with dedicated context windows. Subagent execution wins when packages have clear I/O contracts and procedural instructions; skill-loading grows brittle as task horizon increases (context accumulation); the cost is coordination tokens.
Imported from the existing research archive. Migration preserves the recorded judgment and is not a fresh verification.
What this means for you
This is direct evidence for the architecture choice every agent team faces this year (skill libraries a la Claude Skills vs subagent fleets): encapsulate reusable knowledge as subagents with I/O contracts, and expect the trade to shift further toward subagents as horizons lengthen — at the price of orchestration overhead.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Arxiv | 2609.09233 (Thu 10 Sep digest, announced 2026-09-11T00:00Z) |
|---|---|
| Comparison | skills-as-instructions (main context) vs skills-as-subagents (dedicated context) |
| Findings | subagents win with clear I/O contracts + procedural instructions; skill-loading brittle as horizon grows (context accumulation); cost = coordination tokens |
| Code | not stated in abstract |
Technical impact 3
Engineering value 4
Adoption signal 2
Maturity 2
Verification cost 2
Risk 2