Anthropic's Applied AI team found 'a single agent with skills frequently outperformed' subagent designs across enterprise deployments — and its migration kit on GitHub folds subagents 'one at a time' against the same eval

“Anthropic's applied-AI team found a single agent with skills frequently outperforming subagent designs across enterprise deployments — and shipped a GitHub kit that folds yours one eval at a time.”

Usable claims

• On Thursday, October 22, 2026 (calendar-verified; the page's own 'Date October 22, 2026' metadata row and JSON-LD datePublished 2026-10-22), Anthropic published 'Building agents: When to fold subagents into your main agent loop' by Diane Sarkis on claude.com. The dek: 'When building agents in Claude Platform, using fewer subagents in favor of progressively disclosed context and tools can improve quality while reducing cost and latency. Here’s when and how to migrate.' The framing: 'Subagents – agents with their own context window, tools, and instructions, orchestrated by a coordinator – are a core component of agent design. They are effective at verification, parallel work, and modular permissioning. We use them in many of our own agents that we build using Claude Platform, for example our multi-agent research system.' The new pattern: 'However, recently the Applied AI team has found a pattern of enterprise agent deployments that have benefited from folding routers, planners, and domain subagents into the main agent loop.' The head finding: 'In our comparisons across several enterprise deployments, a single agent with skills frequently outperformed both the one-prompt-for-everything design and the subagent design on quality, and often at a lower cost and latency per task.' The customer example, in the source's own words: 'one company found the dozen of specialist subagents within its customer-facing agent made a consistent on-brand experience difficult to maintain and added latency that hurt the user experience' (the source's own phrasing, quoted as-is). The familiar-progression pattern's feedback: 'most of this structure was doing work the model no longer needed done for it, and that one clear system prompt, 4-5 thin data tools, and firm output guardrails would cover most of the use case.' The boundary: 'To be clear, subagents still have their place, including at the research frontier.' — kept when 'A single agent loop hits measurable performance degradation from context bloat', when 'Multiple tasks need to be tackled in parallel with fresh context windows that can be discarded once the task is done', or when 'A single agent struggles to select appropriate tools or there are conflicting instructions in the system prompt.' The org-chart trap: 'But too often we see enterprises give each team or problem type its own subagent, with the agent architecture becoming a copy of their org chart (Conway’s law at work).' The failure modes: 'Subagent sprawl: Every team ships its own subagent, so companies end up with hundreds of overlapping subagents with no enforced standards and no specific owner.'; 'The Frankenprompt: Every team adds a sentence to the system prompt, resulting in conflicting rules and no consistent voice.'; and latency — 'One customer that evaluated their multi-agent harness against a single agent loop saw a roughly 2x faster runtime with the latter.' The model-capability driver: 'starting with Opus 4.5, models became capable of managing far more in a single loop than most teams expected. You cannot turn off thinking on Opus 5.5, and Fable 5.1, and recent models increasingly narrate their progress and check their work as they go, which makes some harness elements redundant.' and 'in migrations we have helped customers with, rigid multi-agent structures were often what stopped the new model from doing more of the work.' The prompt-cut datum: 'we reduced Claude Code’s system prompt for more advanced models by over 80% which included cuts related to progressively disclosing skills and tools.' The design bar: 'The single agent loop design requires discipline: a carefully managed system prompt, skill and tool descriptions that trigger at the right times, and deliberate handling of context windows, memory, and note-taking.' The counterexample against blind folding: 'When we tested this on a sample support agent, folding one subagent into another lowered latency, but also broke a process that had been passing. The main agent was holding the order lookup tool and answering on its own, without asking the shipping specialist for the delivery date.' — with the eval rule 'run each task more than once, since a break like this one only showed up in one run out of two.' The tool-count threshold: 'once you exceed 30-50 tool definitions in the system prompt, tool choice tends to degrade.' The kit: 'Our migration kit chains these skills together in one guided workflow for assessing and moving multi-agent systems to a single agent loop. It includes a demo you can try before running it on the real thing' — linked to github.com/anthropics/single-agent-loop-kit. The close: 'As models get more capable and agents get longer running, more of your subagents can be skills and tools. You don’t have to make that move in one step, though. Build a small end-to-end eval, fold in one subagent, and keep it as a subagent only if the folded version scores worse, then rinse and repeat.' Per the kit's own README (fetched from the repo's master branch this run): 'A starter kit for moving a multi-agent system (a router or orchestrator in front of specialist subagents) toward one agent loop that loads skills and tools on demand.'; the banner 'Reference implementation. Not maintained and not accepting contributions.'; 'The kit does not assume a single loop always wins. It measures. Every change is scored on the same eval before and after, and a subagent only goes away when the numbers say its replacement holds.'; the steps 'Measure' / 'Combine' / 'Fold' — 'Turn each remaining subagent into a skill and tools in the main agent, one at a time, and compare.'; 'examples/support-desk is a small, runnable support agent with a router and six subagents'; and the no-score line 'Without a score, a migration is a matter of opinion.' Per The Decoder (Matthias Bastian, datePublished 2026-10-25T09:12:10+00:00), the coverage that surfaced the piece to a general audience: 'Anthropic recommends that companies evaluate whether they can consolidate sub-agents into a single agent that loads context on demand.'; 'In one comparison, Anthropic says the single-agent architecture ran about twice as fast as the multi-agent setup.'; and 'Anthropic warns against merging blindly, since in one test with a support agent, folding in a sub-agent did reduce latency but broke a previously working process.'