--- title: "Claim Ledger: Your Multi-Agent System Is an Org Chart" description: "Cognition said don't build them. Anthropic said do. A year on, they converge on the one question that decides it." author: "Harry Floyd" publication: "The Durability Curve" canonical: "https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart/" essay: "https://durabilitycurve.com/blog/your-multi-agent-system-is-an-org-chart/" substack: "https://harryfloyd.substack.com/p/your-multi-agent-system-is-an-org" published: "2026-07-07" ledger_date: "2026-10-03" law: "Law I" claims: 9 removed_before_publication: 0 --- # Claim Ledger: Your Multi-Agent System Is an Org Chart *Cognition said don't build them. Anthropic said do. A year on, they converge on the one question that decides it.* Essay: https://durabilitycurve.com/blog/your-multi-agent-system-is-an-org-chart/ Ledger (canonical, cite this): https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart/ Published 2026-07-07 · last verified 2026-10-03 ## The evidence, row by row Labels: VERIFIED = checked against the primary source itself by a checker who did not draft the piece, usually with the exact words, where they sit and the date read · EXECUTED = a run or observation done for the piece (code, a query, a count, or a check made live in an app); the claim is what it returned; run records are not published · CHECKED = checked against its source by a separate checker in the older ledger format, which recorded no quote, locator or date · REPORTED = not verified word for word against a primary source (secondary, not openable in full, or supported only in part); the essay words it accordingly · STRUCK = drafted, checked and removed before publication · EXCLUDED = considered and deliberately left out. ### MA1. Cognition's June 2025 post, written from its own experience building coding agents and explicitly dated to 2025, said running multiple agents in collaboration produced fragile systems because decision-making ended up too dispersed and context could not be shared thoroughly enough between the agents. - Status: VERIFIED (read 2026-10-03) - Row link: https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart/#c-MA1 - Quote: in 2025, running multiple agents in collaboration only results in fragile systems … decision-making … too dispersed and context isn’t able to be shared thoroughly enough - Source: Walden Yan, Cognition, Don't Build Multi-Agents, 12 June 2025, section "Applying the Principles", sub-section "Multi-Agents", final paragraph - URL: https://cognition.com/blog/dont-build-multi-agents ### MA2. On Anthropic's own internal research eval, not an external benchmark, its multi-agent research system (a Claude Opus 4 lead agent with Claude Sonnet 4 subagents) outperformed single-agent Claude Opus 4 by 90.2%. - Status: VERIFIED (read 2026-10-03) - Row link: https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart/#c-MA2 - Quote: Claude Opus 4 as the lead agent and Claude Sonnet 4 subagents outperformed single-agent Claude Opus 4 by 90.2% on our internal research eval - Source: Anthropic Engineering, How we built our multi-agent research system, 13 June 2025, section "Benefits of a multi-agent system", second paragraph - URL: https://www.anthropic.com/engineering/multi-agent-research-system ### MA3. On the BrowseComp browsing evaluation specifically, Anthropic found that token usage by itself explained 80% of the performance variance; three factors together explained 95%. - Status: VERIFIED (read 2026-10-03) - Row link: https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart/#c-MA3 - Quote: three factors explained 95% of the performance variance in the BrowseComp evaluation … token usage by itself explains 80% of the variance - Source: Anthropic Engineering, How we built our multi-agent research system, 13 June 2025, section "Benefits of a multi-agent system", third paragraph - URL: https://www.anthropic.com/engineering/multi-agent-research-system ### MA4. In Anthropic's own data, multi-agent systems used about 15× more tokens than chats, against about 4× for single agents. - Status: VERIFIED (read 2026-10-03) - Row link: https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart/#c-MA4 - Quote: In our data, agents typically use about 4× more tokens than chat interactions, and multi-agent systems use about 15× more tokens than chats - Source: Anthropic Engineering, How we built our multi-agent research system, 13 June 2025, section "Benefits of a multi-agent system", fourth paragraph - URL: https://www.anthropic.com/engineering/multi-agent-research-system ### MA5. Anthropic wrote in June 2025 that domains requiring all agents to share the same context, or involving many dependencies between agents, are not a good fit for multi-agent systems today, and that most coding tasks involve fewer truly parallelisable tasks than research. - Status: VERIFIED (read 2026-10-03) - Row link: https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart/#c-MA5 - Quote: share the same context or involve many dependencies between agents are not a good fit … most coding tasks involve fewer truly parallelizable tasks than research - Source: Anthropic Engineering, How we built our multi-agent research system, 13 June 2025, section "Benefits of a multi-agent system", fourth paragraph - URL: https://www.anthropic.com/engineering/multi-agent-research-system ### MA6. A preprint by Tran and Kiela found that, with reasoning-token budgets held equal, single-agent systems consistently match or outperform multi-agent systems on multi-hop reasoning tasks across three model families (Qwen3, DeepSeek-R1-Distill-Llama and Gemini 2.5). The finding covers multi-hop reasoning only. - Status: VERIFIED (read 2026-10-03) - Row link: https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart/#c-MA6 - Quote: three model families (Qwen3, DeepSeek-R1-Distill-Llama, and Gemini 2.5) … SAS consistently match or outperform MAS on multi-hop reasoning tasks when reasoning tokens are held constant - Source: Dat Tran and Douwe Kiela, Single-Agent LLMs Outperform Multi-Agent Systems on Multi-Hop Reasoning Under Equal Thinking Token Budgets, arXiv preprint 2604.02460v2, Abstract (revised 11 April 2026) - URL: https://arxiv.org/abs/2604.02460 ### MA7. Claude Code's current sub-agents documentation lists its built-in Explore and Plan subagents with read-only tools, Write and Edit denied. - Status: VERIFIED (read 2026-10-03) - Row link: https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart/#c-MA7 - Quote: Explore … Tools: read-only tools; Write and Edit are denied … Plan … Tools: read-only tools; Write and Edit are denied - Source: Claude Code Docs, Create custom subagents, section "Built-in subagents", Explore and Plan tabs, Tools line - URL: https://code.claude.com/docs/en/sub-agents ### MA8. Cognition's April 2026 follow-up said its original observations still hold for parallel-writer swarms, and that the multi-agent setups that work in its practice are those where multiple agents contribute intelligence while writes stay single-threaded. - Status: VERIFIED (read 2026-10-03) - Row link: https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart/#c-MA8 - Quote: Our original observations still hold today for parallel-writer swarms … setups where multiple agents contribute intelligence to a task while writes stay single-threaded - Source: Walden Yan, Cognition, Multi-Agents: What's Actually Working, 22 April 2026, introduction, second paragraph - URL: https://cognition.com/blog/multi-agents-working ### MA9. Cognition states, as its own company figure, that Devin usage in its largest-enterprises segment, which it describes as traditionally cautious toward new technologies, grew roughly 8× over the six months before its April 2026 post. - Status: VERIFIED (read 2026-10-03) - Row link: https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart/#c-MA9 - Quote: Devin usage in our largest enterprises segment … traditionally been cautious toward adopting new technologies, we see an explosion over the last 6 months (~8x) - Source: Walden Yan, Cognition, Multi-Agents: What's Actually Working, 22 April 2026, section "What Changed in the Last 10 Months", first paragraph - URL: https://cognition.com/blog/multi-agents-working ## Cite - A claim: name the original source (and its locator) first, then the row it was checked in, e.g. ", . Checked in: Claim Ledger, "Your Multi-Agent System Is an Org Chart", The Durability Curve, row , https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart/#c-." - The essay: Floyd, Harry (2026). Your Multi-Agent System Is an Org Chart. The Durability Curve. https://durabilitycurve.com/blog/your-multi-agent-system-is-an-org-chart/ - This ledger: Floyd, Harry (2026). Claim Ledger: Your Multi-Agent System Is an Org Chart. The Durability Curve. https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart/ - Downloads: https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart.csv · https://durabilitycurve.com/claims/your-multi-agent-system-is-an-org-chart.json Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Reuse it, quote it or train on it, with credit to The Durability Curve and a link. Open to AI (robots.txt: search=yes, ai-input=yes, ai-train=yes).