{"id":"e3071000eb21c8f4cc98cec39e0bbc556b82760d3c8902f8b2494d689836f90c","title":"Anthropic reports domain expertise shapes how people use Claude Code","argument":"Anthropic’s report, published June 16, 2026, examines how people use Claude Code through an analysis of interactive sessions. It argues that people usually decide what work to do while Claude handles much of how to do it. Sessions rated as more expert tend to set off more work per instruction and are more likely to succeed under the report’s transcript-based measures. The report treats expertise as task-specific, not simply a matter of job title or coding background.\n\nAnthropic also reports a shift toward operating software, data analysis and writing, alongside a fall in the share of sessions spent fixing code. It estimates that task value rose, comparing work with freelance job postings, but cautions that these estimates are coarse. The authors argue that coding agents may absorb some implementation-heavy work while rewarding workers who understand the problems they are solving. They conclude that a working grasp of a domain captures most of the benefit, while deep specialization adds less. The findings are preliminary: the authors cannot measure whether session outputs are used in the real world, and the report excludes non-interactive usage.","topicIds":["work-and-business--jobs-and-occupations","job-displacement-and-creation"],"domain":"other","claimType":"attributed reading","limits":"Text as retrieved on October 28, 2026; originally published June 16, 2026. The authors say the findings are preliminary, cannot measure real-world outcomes such as whether code is used or discarded, exclude a substantial share of non-interactive usage, and rely on transcript-reading model classifications that are challenging to validate at scale.","speaker":{"name":"Anthropic","roleAtSource":"Publisher"},"source":{"title":"How Claude Code is used in practice","publisher":"Anthropic","format":"Article","url":"https://www.anthropic.com/research/claude-code-expertise","publishedAt":"2026-06-16T00:00:00.000Z","dateBasis":"Date stated on the publisher’s page."},"evidence":[{"text":"Anthropic reports that users make most planning decisions while Claude makes most execution decisions in a typical session.","excerpt":"On average, people make about 70% of the planning decisions but only 20% of the execution decisions","url":"https://www.anthropic.com/research/claude-code-expertise","relationship":"Source-stated"},{"text":"Anthropic reports that greater task-specific expertise is associated with more Claude activity per prompt.","excerpt":"expert sessions set off action chains more than twice as long (12 actions) carrying five times the output (3,200 words)","url":"https://www.anthropic.com/research/claude-code-expertise","relationship":"Source-stated"},{"text":"Anthropic reports that sessions rated as more expert are more likely to succeed under its transcript-based measures.","excerpt":"A novice-rated session reaches our strictest measure, verified success, 15% of the time and at least partial success 77% of the time.","url":"https://www.anthropic.com/research/claude-code-expertise","relationship":"Source-stated"},{"text":"Anthropic estimates that the average session’s value rose, while cautioning that its task-value estimates are coarse.","excerpt":"The rise holds across many kinds of work. Building, operating, and fixing-type tasks all grew more valuable by roughly a third or more (about 43%, 34%, and 32% respectively). These price estimates are coarse","url":"https://www.anthropic.com/research/claude-code-expertise","relationship":"Source-stated"},{"text":"Anthropic argues that a working grasp of a domain captures most of the benefit, while deep specialization adds less.","excerpt":"a working grasp of the domain captures most of the benefit, while deep specialization adds only a bit more beyond that.","url":"https://www.anthropic.com/research/claude-code-expertise","relationship":"Source-stated"},{"text":"Anthropic says it cannot measure real-world outcomes such as whether code produced in a session is later used or discarded.","excerpt":"we cannot measure real-world outcomes, like whether code written in a session is actually used or discarded thereafter","url":"https://www.anthropic.com/research/claude-code-expertise","relationship":"Source-stated"}],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.anthropic.com/research/claude-code-expertise","checkedAt":null,"items":[]},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Do these patterns hold across the economy, including in non-interactive use and in real-world outcomes the report could not measure?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Anthropic reports domain expertise shapes how people use Claude Code","summary":"Anthropic’s report, published June 16, 2026, examines how people use Claude Code through an analysis of interactive sessions. It argues that people usually decide what work to do while Claude handles much of how to do it. Sessions rated as more expert tend to set off more work per instruction and are more likely to succeed under the report’s transcript-based measures. The report treats expertise as task-specific, not simply a matter of job title or coding background.\n\nAnthropic also reports a shift toward operating software, data analysis and writing, alongside a fall in the share of sessions spent fixing code. It estimates that task value rose, comparing work with freelance job postings, but cautions that these estimates are coarse. The authors argue that coding agents may absorb some implementation-heavy work while rewarding workers who understand the problems they are solving. They conclude that a working grasp of a domain captures most of the benefit, while deep specialization adds less. The findings are preliminary: the authors cannot measure whether session outputs are used in the real world, and the report excludes non-interactive usage.","question":"Do these patterns hold across the economy, including in non-interactive use and in real-world outcomes the report could not measure?","claims":[{"text":"Anthropic reports that users make most planning decisions while Claude makes most execution decisions in a typical session.","excerpt":"On average, people make about 70% of the planning decisions but only 20% of the execution decisions","support":"source-stated"},{"text":"Anthropic reports that greater task-specific expertise is associated with more Claude activity per prompt.","excerpt":"expert sessions set off action chains more than twice as long (12 actions) carrying five times the output (3,200 words)","support":"source-stated"},{"text":"Anthropic reports that sessions rated as more expert are more likely to succeed under its transcript-based measures.","excerpt":"A novice-rated session reaches our strictest measure, verified success, 15% of the time and at least partial success 77% of the time.","support":"source-stated"},{"text":"Anthropic estimates that the average session’s value rose, while cautioning that its task-value estimates are coarse.","excerpt":"The rise holds across many kinds of work. Building, operating, and fixing-type tasks all grew more valuable by roughly a third or more (about 43%, 34%, and 32% respectively). These price estimates are coarse","support":"source-stated"},{"text":"Anthropic argues that a working grasp of a domain captures most of the benefit, while deep specialization adds less.","excerpt":"a working grasp of the domain captures most of the benefit, while deep specialization adds only a bit more beyond that.","support":"source-stated"},{"text":"Anthropic says it cannot measure real-world outcomes such as whether code produced in a session is later used or discarded.","excerpt":"we cannot measure real-world outcomes, like whether code written in a session is actually used or discarded thereafter","support":"source-stated"}],"limitations":["Text as retrieved on October 28, 2026; originally published June 16, 2026.","The authors say the findings are preliminary, cannot measure real-world outcomes such as whether code is used or discarded, exclude a substantial share of non-interactive usage, and rely on transcript-reading model classifications that are challenging to validate at scale."],"mediaId":null,"correctionReason":null,"id":"e3071000eb21c8f4cc98cec39e0bbc556b82760d3c8902f8b2494d689836f90c","publicationId":"852e4d92-9f66-2246-bf6f-114380112e68","version":1,"sourceUrl":"https://www.anthropic.com/research/claude-code-expertise","sourceTitle":"How Claude Code is used in practice","publisher":"Anthropic","publishedAt":"2026-06-16T00:00:00.000Z","originalPublished":{"value":"2026-06-16T00:00:00.000Z","precision":"instant","basis":"backfill:article_published"},"firstObservedAt":"2026-10-20T05:21:31.623Z","createdAt":"2026-11-06T14:03:23.427Z","historical":true,"entityIds":["anthropic"],"topicIds":["work-and-business--jobs-and-occupations","job-displacement-and-creation"],"media":[],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.anthropic.com/research/claude-code-expertise","checkedAt":null,"items":[]},"editorialDecision":{"outcome":"accept","mediaId":null,"inputHash":"f7df6d9a4f085ec142791c5b7b639e31add8313f823ecd6a1da3649e1fdf8340","rubricVersion":"unrecorded","provider":"typesafe","model":"jev-1.13.0","reasonCodes":["bridge_admitted_by_model"],"confidence":0.83,"decidedAt":"2026-10-28T02:12:11.764Z"},"provenance":{"textHash":"a349b67ddab25c30bb26cc2733959a9f5087862eb6bd0cdd3bee17ec1364fb0f","observationId":"5c53f6ba76c0fc604b597bcd6c70458cbc939ef6e34ae3c6a5bc7d5257917e9e","retrievedAt":"2026-10-28T02:13:52.557Z","kind":"publisher-page","preparer":{"model":"openai/gpt-6-luna","runId":"1e2d9144-ecf8-b960-7bdb-3adb4c767abd"},"reviewer":{"model":"anthropic/claude-opus-5.5","runId":"415583be-a1a7-00fa-fa3e-1cdbdd702da4"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-06T13:58:26.966Z","model":"anthropic/claude-opus-5.5"}},"canonicalUrl":"https://escapemeridian.com/readings/anthropic-reports-domain-expertise-shapes-how-people-use-claude-e3071000","formats":{"html":"/readings/anthropic-reports-domain-expertise-shapes-how-people-use-claude-e3071000","markdown":"/readings/anthropic-reports-domain-expertise-shapes-how-people-use-claude-e3071000.md","json":"/readings/anthropic-reports-domain-expertise-shapes-how-people-use-claude-e3071000.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[],"entities":[{"id":"anthropic","name":"Anthropic","kind":"organization","context":"Anthropic develops Claude and publishes research about AI capability, safety and how people use its systems.","referenceUrls":["https://www.anthropic.com/news","https://www.anthropic.com/research/team/economics"],"path":"/entities/anthropic","json":"/entities/anthropic.json","markdown":"/entities/anthropic.md"},{"id":"claude","name":"Claude","kind":"model","context":"Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding.","referenceUrls":["https://claude.com/product/overview","https://www.anthropic.com/news","https://www.anthropic.com/research/team/economics"],"path":"/entities/claude","json":"/entities/claude.json","markdown":"/entities/claude.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"e3071000eb21c8f4cc98cec39e0bbc556b82760d3c8902f8b2494d689836f90c","revision":"852e4d92-9f66-2246-bf6f-114380112e68","revisionKind":"publication-id","title":"Anthropic reports domain expertise shapes how people use Claude Code","url":"https://escapemeridian.com/readings/anthropic-reports-domain-expertise-shapes-how-people-use-claude-e3071000"},"source":{"type":"source","id":"https://www.anthropic.com/research/claude-code-expertise","revision":"a349b67ddab25c30bb26cc2733959a9f5087862eb6bd0cdd3bee17ec1364fb0f","revisionKind":"sha256-source-text","title":"How Claude Code is used in practice","url":"https://www.anthropic.com/research/claude-code-expertise"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter."},"revisions":[{"id":"e3071000eb21c8f4cc98cec39e0bbc556b82760d3c8902f8b2494d689836f90c","title":"Anthropic reports domain expertise shapes how people use Claude Code","argument":"Anthropic’s report, published June 16, 2026, examines how people use Claude Code through an analysis of interactive sessions. It argues that people usually decide what work to do while Claude handles much of how to do it. Sessions rated as more expert tend to set off more work per instruction and are more likely to succeed under the report’s transcript-based measures. The report treats expertise as task-specific, not simply a matter of job title or coding background.\n\nAnthropic also reports a shift toward operating software, data analysis and writing, alongside a fall in the share of sessions spent fixing code. It estimates that task value rose, comparing work with freelance job postings, but cautions that these estimates are coarse. The authors argue that coding agents may absorb some implementation-heavy work while rewarding workers who understand the problems they are solving. They conclude that a working grasp of a domain captures most of the benefit, while deep specialization adds less. The findings are preliminary: the authors cannot measure whether session outputs are used in the real world, and the report excludes non-interactive usage.","topicIds":["work-and-business--jobs-and-occupations","job-displacement-and-creation"],"domain":"other","claimType":"attributed reading","limits":"Text as retrieved on October 28, 2026; originally published June 16, 2026. The authors say the findings are preliminary, cannot measure real-world outcomes such as whether code is used or discarded, exclude a substantial share of non-interactive usage, and rely on transcript-reading model classifications that are challenging to validate at scale.","speaker":{"name":"Anthropic","roleAtSource":"Publisher"},"source":{"title":"How Claude Code is used in practice","publisher":"Anthropic","format":"Article","url":"https://www.anthropic.com/research/claude-code-expertise","publishedAt":"2026-06-16T00:00:00.000Z","dateBasis":"Date stated on the publisher’s page."},"evidence":[{"text":"Anthropic reports that users make most planning decisions while Claude makes most execution decisions in a typical session.","excerpt":"On average, people make about 70% of the planning decisions but only 20% of the execution decisions","url":"https://www.anthropic.com/research/claude-code-expertise","relationship":"Source-stated"},{"text":"Anthropic reports that greater task-specific expertise is associated with more Claude activity per prompt.","excerpt":"expert sessions set off action chains more than twice as long (12 actions) carrying five times the output (3,200 words)","url":"https://www.anthropic.com/research/claude-code-expertise","relationship":"Source-stated"},{"text":"Anthropic reports that sessions rated as more expert are more likely to succeed under its transcript-based measures.","excerpt":"A novice-rated session reaches our strictest measure, verified success, 15% of the time and at least partial success 77% of the time.","url":"https://www.anthropic.com/research/claude-code-expertise","relationship":"Source-stated"},{"text":"Anthropic estimates that the average session’s value rose, while cautioning that its task-value estimates are coarse.","excerpt":"The rise holds across many kinds of work. Building, operating, and fixing-type tasks all grew more valuable by roughly a third or more (about 43%, 34%, and 32% respectively). These price estimates are coarse","url":"https://www.anthropic.com/research/claude-code-expertise","relationship":"Source-stated"},{"text":"Anthropic argues that a working grasp of a domain captures most of the benefit, while deep specialization adds less.","excerpt":"a working grasp of the domain captures most of the benefit, while deep specialization adds only a bit more beyond that.","url":"https://www.anthropic.com/research/claude-code-expertise","relationship":"Source-stated"},{"text":"Anthropic says it cannot measure real-world outcomes such as whether code produced in a session is later used or discarded.","excerpt":"we cannot measure real-world outcomes, like whether code written in a session is actually used or discarded thereafter","url":"https://www.anthropic.com/research/claude-code-expertise","relationship":"Source-stated"}],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.anthropic.com/research/claude-code-expertise","checkedAt":null,"items":[]},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Do these patterns hold across the economy, including in non-interactive use and in real-world outcomes the report could not measure?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Anthropic reports domain expertise shapes how people use Claude Code","summary":"Anthropic’s report, published June 16, 2026, examines how people use Claude Code through an analysis of interactive sessions. It argues that people usually decide what work to do while Claude handles much of how to do it. Sessions rated as more expert tend to set off more work per instruction and are more likely to succeed under the report’s transcript-based measures. The report treats expertise as task-specific, not simply a matter of job title or coding background.\n\nAnthropic also reports a shift toward operating software, data analysis and writing, alongside a fall in the share of sessions spent fixing code. It estimates that task value rose, comparing work with freelance job postings, but cautions that these estimates are coarse. The authors argue that coding agents may absorb some implementation-heavy work while rewarding workers who understand the problems they are solving. They conclude that a working grasp of a domain captures most of the benefit, while deep specialization adds less. The findings are preliminary: the authors cannot measure whether session outputs are used in the real world, and the report excludes non-interactive usage.","question":"Do these patterns hold across the economy, including in non-interactive use and in real-world outcomes the report could not measure?","claims":[{"text":"Anthropic reports that users make most planning decisions while Claude makes most execution decisions in a typical session.","excerpt":"On average, people make about 70% of the planning decisions but only 20% of the execution decisions","support":"source-stated"},{"text":"Anthropic reports that greater task-specific expertise is associated with more Claude activity per prompt.","excerpt":"expert sessions set off action chains more than twice as long (12 actions) carrying five times the output (3,200 words)","support":"source-stated"},{"text":"Anthropic reports that sessions rated as more expert are more likely to succeed under its transcript-based measures.","excerpt":"A novice-rated session reaches our strictest measure, verified success, 15% of the time and at least partial success 77% of the time.","support":"source-stated"},{"text":"Anthropic estimates that the average session’s value rose, while cautioning that its task-value estimates are coarse.","excerpt":"The rise holds across many kinds of work. Building, operating, and fixing-type tasks all grew more valuable by roughly a third or more (about 43%, 34%, and 32% respectively). These price estimates are coarse","support":"source-stated"},{"text":"Anthropic argues that a working grasp of a domain captures most of the benefit, while deep specialization adds less.","excerpt":"a working grasp of the domain captures most of the benefit, while deep specialization adds only a bit more beyond that.","support":"source-stated"},{"text":"Anthropic says it cannot measure real-world outcomes such as whether code produced in a session is later used or discarded.","excerpt":"we cannot measure real-world outcomes, like whether code written in a session is actually used or discarded thereafter","support":"source-stated"}],"limitations":["Text as retrieved on October 28, 2026; originally published June 16, 2026.","The authors say the findings are preliminary, cannot measure real-world outcomes such as whether code is used or discarded, exclude a substantial share of non-interactive usage, and rely on transcript-reading model classifications that are challenging to validate at scale."],"mediaId":null,"correctionReason":null,"id":"e3071000eb21c8f4cc98cec39e0bbc556b82760d3c8902f8b2494d689836f90c","publicationId":"852e4d92-9f66-2246-bf6f-114380112e68","version":1,"sourceUrl":"https://www.anthropic.com/research/claude-code-expertise","sourceTitle":"How Claude Code is used in practice","publisher":"Anthropic","publishedAt":"2026-06-16T00:00:00.000Z","originalPublished":{"value":"2026-06-16T00:00:00.000Z","precision":"instant","basis":"backfill:article_published"},"firstObservedAt":"2026-10-20T05:21:31.623Z","createdAt":"2026-11-06T14:03:23.427Z","historical":true,"entityIds":["anthropic"],"topicIds":["work-and-business--jobs-and-occupations","job-displacement-and-creation"],"media":[],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.anthropic.com/research/claude-code-expertise","checkedAt":null,"items":[]},"editorialDecision":{"outcome":"accept","mediaId":null,"inputHash":"f7df6d9a4f085ec142791c5b7b639e31add8313f823ecd6a1da3649e1fdf8340","rubricVersion":"unrecorded","provider":"typesafe","model":"jev-1.13.0","reasonCodes":["bridge_admitted_by_model"],"confidence":0.83,"decidedAt":"2026-10-28T02:12:11.764Z"},"provenance":{"textHash":"a349b67ddab25c30bb26cc2733959a9f5087862eb6bd0cdd3bee17ec1364fb0f","observationId":"5c53f6ba76c0fc604b597bcd6c70458cbc939ef6e34ae3c6a5bc7d5257917e9e","retrievedAt":"2026-10-28T02:13:52.557Z","kind":"publisher-page","preparer":{"model":"openai/gpt-6-luna","runId":"1e2d9144-ecf8-b960-7bdb-3adb4c767abd"},"reviewer":{"model":"anthropic/claude-opus-5.5","runId":"415583be-a1a7-00fa-fa3e-1cdbdd702da4"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-06T13:58:26.966Z","model":"anthropic/claude-opus-5.5"}}}],"audits":[{"publicationId":"852e4d92-9f66-2246-bf6f-114380112e68","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading attributes the analysis, findings and arguments to Anthropic, the publisher of this single research report. Source claims stay framed as what Anthropic reports, argues or estimates.","excerpt":"This report provides evidence on how Claude Code is used in practice, based on a privacy-preserving analysis of ~400,000 interactive sessions"},{"dimension":"summary","status":"uncertain","explanation":"The summary is faithful, but its central term is never explained for a general reader. It does not say that Claude Code is Anthropic's AI coding agent, which takes actions such as editing files and running commands at a person's direction. No definitions are supplied. Interactive versus non-interactive usage is also left unexplained. The reading omits the headline occupation finding that non-software occupations succeed at near software-engineer rates on coding tasks. That omission is less critical. Smallest repair: add a plain-language clause defining Claude Code as an agentic coding tool.","excerpt":"On coding tasks, every major occupation succeeds––accomplishes what the person set out to do, with verifiable evidence like passing tests or committed work––at nearly the same rate as software engineers, on average."},{"dimension":"date","status":"pass","explanation":"The reading displays June 16, 2026, which matches the UTC day of originalPublished. That date is consistent with the data window of October 2025 to April 2026. The page was modified on 2026-10-23, but the reading relies only on core report text, with no later-event or hindsight framing. The limitations note the retrieval date. Forecasts are kept conditional. The kind is publisher-page, so no archive disclosure is needed.","excerpt":"~400,000 interactive sessions from ~235,000 people between October 2025 and April 2026"},{"dimension":"support","status":"pass","explanation":"Each claim's excerpt is verbatim and supports it. The excerpts cover planning versus execution shares, activity per prompt by expertise, the success gradient, coarse value estimates and working grasp versus mastery. The limitations claim is also supported. The question echoes the source's own conditional. The page was edited after publication, but the cited passages are core findings plainly part of the report. The source's conditional framing of labor-market implications is preserved.","excerpt":"If these patterns hold across the economy, it suggests that while agentic coding tools may be absorbing some implementation-heavy work, they are also rewarding those with firm understanding of the problems they solve on the job."},{"dimension":"media","status":"pass","explanation":"No media candidates were supplied and none was selected, so there is no media identity or relevance issue.","excerpt":null}],"createdAt":"2026-11-06T13:58:26.966Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}