Breaking

Anthropic Urges Teams to Reassess Subagents in Favor of Skills and Tools

October 25, 2026 at 07:46 EDT

AI Agents

Anthropic is advising developers to fold some specialist subagents into a main agent loop that loads skills and tools when needed. The company says one customer’s single-loop setup ran roughly twice as fast as its multi-agent system, while stressing that subagents still suit parallel work, verification and other specific needs.

October 22 · Anthropic

Fewer subagents, more shared context

Anthropic recommends evaluating whether specialist prompts can become skills and API calls can become tools in one main agent loop. Subagents still fit parallel work and other measured needs.

A customer comparison

Reported runtime, shown as relative multiples

~2x

Multi-agent runtime

Single-loop runtime

One customer’s reported comparison; not a universal benchmark.

Fold into the loop

Use skills for team procedures and tools for actions or live system data. Evaluate one change at a time.

Keep specialists when needed

Parallel tasks, context limits, tool-selection problems or conflicting instructions can justify subagents.

Measure before migrating: Anthropic suggests starting with an end-to-end evaluation set, folding in one subagent at a time and checking for task failures as well as speed.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.