Skip to content
Escape Meridian

· Apr 28, 2025

Anthropic reports how Claude is used for coding

Markdown · JSON · Original source

published this analysis on April 28, 2025, comparing coding-related conversations on .ai and Claude Code. It argues that specialist coding agents may involve more task automation: 79% of Claude Code conversations were classified as automation, compared with 49% on Claude.ai. But Anthropic says “Feedback Loop” interactions still require user input, such as returning error messages, so automation does not mean people are absent from the work. Anthropic also found frequent use for user-facing apps and interfaces. It suggests jobs focused on simple applications and interfaces might face earlier disruption if “vibe coding”—describing desired outcomes in natural language and letting AI handle implementation details—moves into mainstream workflows. The source’s conclusion is that software development may offer an early indication of how work could change, while its lessons may not carry over to other occupations. Anthropic asks whether developers will remain involved through feedback loops or shift toward managing AI systems, and which software roles might change or disappear. Its analysis describes Claude use, not outcomes across the developer workforce.

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.

AnthropicArticle · anthropic.comOpen original source

Original source

· Article

Source details

Original source date: Apr 28, 2025 (2025-04-28T00:00:00.000Z).

Date basis: Date stated on the publisher’s page.

Added to Escape Meridian: 2026-11-14T12:18:23.594Z. First observed: 2026-10-18T09:41:19.813Z. Retrieved: 2026-10-27T22:05:19.351Z.

Referenced by

Recorded citations and public file attachments. A reference alone does not establish agreement or influence.

No public references recorded here yet.

Reference data (JSON)

Limits

Text as retrieved on October 27, 2026; originally published April 28, 2025. analyzed .ai and Claude Code only, excluding Team, Enterprise, and usage. Anthropic says it did not study how developers ultimately used AI outputs in their codebases, the resulting code quality, or whether interactions improved productivity or code quality. Anthropic says its project-type classifications relied on limited context and educated guesses, and that some classifications might be incorrect. The source’s claims describe Claude usage and do not establish effects across the broader developer population.

Open question

As AI capabilities advance, will developers continue to guide work through feedback loops, or shift toward mostly managing AI systems, as asks?

Source-stated

Anthropic reports that automation was more common in Claude Code conversations than in Claude.ai conversations.

79% of conversations on Claude Code were identified as “automation”—where AI directly performs tasks—rather than “augmentation,” where AI collaborates with and enhances human capabilities (21%).

Source-stated

Anthropic says feedback-loop interactions still involve human input, such as returning error messages to Claude.

“Feedback Loop” interactions still require user input (even if that input is simply pasting error messages back to Claude).

Source-stated

Anthropic suggests jobs focused on simple applications and user interfaces might face earlier disruption if vibe coding enters mainstream workflows.

jobs that center on making simple applications and user interfaces might face earlier disruption from AI systems if increasing capabilities cause “vibe coding” to shift more into mainstream workflows.

Source-stated

Anthropic describes its findings about which groups use Claude for coding as preliminary because the project categories rely on uncertain inferences from incomplete data.

Because we don’t know the real-world context in which Claude’s responses were being used, these analyses rely on uncertain inferences from incomplete data. We therefore treat these findings as more preliminary than the ones described above.

Source-stated

Anthropic says software development may offer information about how other occupations could change, while cautioning that those lessons may not carry over directly.

Although we can’t assume that the lessons we draw from software development will directly carry over to other types of occupation, software development might be a leading indicator that gives us useful information about how other occupations might change

Source and contributor context

Anthropic · Publisher

The publisher is identified here. No individual author has been inferred.

Source media

Media retrieval status: unchecked.

Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.

API

An application programming interface: a defined way for software to request data or actions from another system.

Anthropic

Anthropic develops Claude and publishes research about AI capability, safety and how people use its systems.

Claude

Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding.

Editorial provenance

Version 1. Publication ID: 5894c005-5c9a-e9f4-2e96-4d396c72d5d0.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: passed by anthropic/claude-opus-5.5 on 2026-11-14T12:13:44.555Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge_admitted_by_model.

Observation: 5d372d69fbd6c8a4e0aceab718c8aad5d0edf71c223365e3be4ad4d84a53482d. Source text hash: 2a454b398245f3949874c0a1988285da30fcf038cc921aea584c3c31090c8a09.

Revision and source-check history

Version 1 · 2026-11-14T12:18:23.594Z

Anthropic reports how Claude is used for coding

published this analysis on April 28, 2025, comparing coding-related conversations on .ai and Claude Code. It argues that specialist coding agents may involve more task automation: 79% of Claude Code conversations were classified as automation, compared with 49% on Claude.ai. But Anthropic says “Feedback Loop” interactions still require user input, such as returning error messages, so automation does not mean people are absent from the work. Anthropic also found frequent use for user-facing apps and interfaces. It suggests jobs focused on simple applications and interfaces might face earlier disruption if “vibe coding”—describing desired outcomes in natural language and letting AI handle implementation details—moves into mainstream workflows. The source’s conclusion is that software development may offer an early indication of how work could change, while its lessons may not carry over to other occupations. Anthropic asks whether developers will remain involved through feedback loops or shift toward managing AI systems, and which software roles might change or disappear. Its analysis describes Claude use, not outcomes across the developer workforce.

Original source

· Article

Source details

Original source date: Apr 28, 2025 (2025-04-28T00:00:00.000Z).

Date basis: Date stated on the publisher’s page.

Added to Escape Meridian: 2026-11-14T12:18:23.594Z. First observed: 2026-10-18T09:41:19.813Z. Retrieved: 2026-10-27T22:05:19.351Z.

Limits

Text as retrieved on October 27, 2026; originally published April 28, 2025. analyzed .ai and Claude Code only, excluding Team, Enterprise, and usage. Anthropic says it did not study how developers ultimately used AI outputs in their codebases, the resulting code quality, or whether interactions improved productivity or code quality. Anthropic says its project-type classifications relied on limited context and educated guesses, and that some classifications might be incorrect. The source’s claims describe Claude usage and do not establish effects across the broader developer population.

Open question

As AI capabilities advance, will developers continue to guide work through feedback loops, or shift toward mostly managing AI systems, as asks?

Source-stated

Anthropic reports that automation was more common in Claude Code conversations than in Claude.ai conversations.

79% of conversations on Claude Code were identified as “automation”—where AI directly performs tasks—rather than “augmentation,” where AI collaborates with and enhances human capabilities (21%).

Source-stated

Anthropic says feedback-loop interactions still involve human input, such as returning error messages to Claude.

“Feedback Loop” interactions still require user input (even if that input is simply pasting error messages back to Claude).

Source-stated

Anthropic suggests jobs focused on simple applications and user interfaces might face earlier disruption if vibe coding enters mainstream workflows.

jobs that center on making simple applications and user interfaces might face earlier disruption from AI systems if increasing capabilities cause “vibe coding” to shift more into mainstream workflows.

Source-stated

Anthropic describes its findings about which groups use Claude for coding as preliminary because the project categories rely on uncertain inferences from incomplete data.

Because we don’t know the real-world context in which Claude’s responses were being used, these analyses rely on uncertain inferences from incomplete data. We therefore treat these findings as more preliminary than the ones described above.

Source-stated

Anthropic says software development may offer information about how other occupations could change, while cautioning that those lessons may not carry over directly.

Although we can’t assume that the lessons we draw from software development will directly carry over to other types of occupation, software development might be a leading indicator that gives us useful information about how other occupations might change

Source and contributor context

Anthropic · Publisher

The publisher is identified here. No individual author has been inferred.

Source media

Media retrieval status: unchecked.

Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.

API

An application programming interface: a defined way for software to request data or actions from another system.

Anthropic

Anthropic develops Claude and publishes research about AI capability, safety and how people use its systems.

Claude

Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding.

Editorial provenance

Version 1. Publication ID: 5894c005-5c9a-e9f4-2e96-4d396c72d5d0.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: passed by anthropic/claude-opus-5.5 on 2026-11-14T12:13:44.555Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge_admitted_by_model.

Observation: 5d372d69fbd6c8a4e0aceab718c8aad5d0edf71c223365e3be4ad4d84a53482d. Source text hash: 2a454b398245f3949874c0a1988285da30fcf038cc921aea584c3c31090c8a09.

passed · 2026-11-14T12:13:44.555Z · anthropic/claude-opus-5.5

Publication: 5894c005-5c9a-e9f4-2e96-4d396c72d5d0

attribution — pass: This is one Anthropic research article, not a listing. The reading consistently attributes its findings, suggestions and questions to Anthropic and does not present them as established fact or as Escape Meridian's view.

In our previous Economic Index research, we found very disproportionate use of Claude by US workers in computer-related occupations

summary — pass: The reading leads with the actor and the comparison, gives the 79% vs 49% automation split, and explains Feedback Loop input and vibe coding in plain language. It also keeps the hedged disruption claim, the caution about other occupations, and the source's open questions. It describes Claude Code as a specialist coding agent only by implication, but a general reader can still follow the reading. Leaving out the startup finding does not distort the source's meaning.

This raises questions about the extent to which developers will still be involved as AI use becomes more common.

date — pass: April 28, 2025 matches the UTC calendar day of originalPublished. The page was modified in October 2026, but every claim the reading relies on comes from the article's core findings and limitations, which are plainly part of the original piece. The reading uses no hindsight framing and reports Anthropic's forecasts as its own conditional suggestions. This is a publisher page, so no archive disclosure is needed.

support — pass: Every claim excerpt is verbatim and supports its claim. The disruption point stays conditional, and the question combines two questions Anthropic itself raises. The limitations match the source's own list. The reading claims no unsupported benefits, costs or outcomes, and it notes that the findings cover Claude use only.

We only studied what developers delegate to AI—not how they ultimately use AI outputs in their codebase, the quality of the resulting code, or whether these interactions effectively improved productivity or code quality.

media — pass: No media is selected and no verified media candidates were supplied, so nothing was invented or misidentified.

Challenge or correct this reading

Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.