A privacy-preserving analysis of ~400,000 Claude Code sessions from ~235,000 people (October 2025 to April 2026). People decide what to build, the agent decides how to build it. Whether a session succeeds turns more on how well you understand the problem than on whether you can code. Two numbers frame the study.
Each session is classified into the single work mode that best describes what it is trying to accomplish. About 56% of sessions involve writing, fixing, or testing code directly; 17% operate software; 14% plan or explore; and 13% produce analysis or prose.
A decision-attribution classifier separates every meaningful decision in a session into planning (what to do, which approach, what counts as done) and execution (which files to change, what code to write, which commands to run), then attributes each to the user or to Claude.
Median: users make about 70% of planning decisions but only about 20% of execution decisions. The "how" is largely delegated to the agent.
A typical session runs about four turns. Each prompt sets off a chain of around 10 Claude actions on average (sometimes over 100), producing about 2,400 words per turn. When the user keeps execution control (>80% of execution decisions), Claude takes about 8 actions per turn; when Claude takes over planning (>80%), it takes about 16.
People decide what to build, and the agent decides how to build it.— Anthropic, "Agentic coding and persistent returns to expertise" "
A classifier rates each user's apparent expertise at the task on a five-point scale, from novice to expert. It looks at three signals: how precisely the user frames their directions, what they ask Claude to verify, and whether the user tends to correct Claude or Claude tends to correct the user. This expertise is task-specific, distinct from job title or general ability.
A senior engineer asking their first Rust question is a beginner at Rust.
An accountant who has never used Python, but tells Claude exactly which reconciliation rules a script must enforce and catches the edge case it mishandles at month-end close, is an expert at that task.
Controlling for work mode, task value, month, occupation, and model family, the trend remains significant: about +9% actions and +13% output per expertise level (p < 0.001). The gap appears within every kind of work and every band of task value.
Success is measured as judged success (classifier reads the full transcript) and the stricter verified success (judged successful plus at least one hard signal: matching git commits, passing tests, or explicit user affirmation). Across all measures, more expertise means more success. Most of the gain comes from novice to intermediate; the slope flattens between intermediate and expert.
Abandoned rate (judged failed, zero lines written): novice 19% vs everyone else 5-7%. The least experienced users give up at several times the rate of everyone else when struggling.
The study infers occupation from transcripts using the SOC taxonomy and explicitly instructs the classifier not to treat coding as evidence of a coding profession. A lawyer who builds a script to flag missing clauses is mapped to Legal Occupations.
The composition of work changed substantially over the seven months. The share of sessions spent fixing broken code fell by nearly half; the freed-up share went to operating software, writing, and data analysis.
Tasks also grew more valuable. Approximated by what the work would cost on a freelance marketplace, the average session value rose about 27% (the Key Findings summarize this as "about 25% on average"); building +43%, operating +34%, fixing +32%. The study notes these are coarse estimates meant for relative comparison over time.
The first six sections are evidence from inside sessions — Anthropic's 400,000 sessions. The job market is a separate source. In a July 2026 analysis, Indeed's Hiring Lab observed that software development job postings have rebounded, and that the rebound is concentrated almost entirely in senior, AI-fluent roles. The two sources use different lenses (session-level success vs. labor-market demand), but point at the same thing: value accrues to the senior, domain-fluent people.
Software development postings rose almost 15% while overall postings fell 7% over the same period. The rebound coincides with the arrival of Claude Code and other agentic AI tools, but Hiring Lab is explicit that correlation does not imply causation.
Between May 2025 and May 2026, 71% of the increase in software development postings came from senior roles, and 37% came from jobs that mention AI in their title (the two categories overlap). Demand is growing for experienced people who can work with AI, not a broad-based recovery across all software roles. Hiring Lab frames this as possible seniority-biased technological change, in line with earlier research on AI's disproportionate impact on entry-level jobs.
Two pieces of context must be read alongside this, or the recovery looks larger than it is. First, the rebound starts from a low base: even after rising almost 15%, software development postings remain about 27.5% below their pre-pandemic level, while overall postings are essentially back to February 2020. Second, the relationship between AI exposure and postings flipped over these years: from 2022 to 2026, the more AI-exposed an occupation (including software development), the more it declined; from 2025 to 2026, the more AI-exposed, the more it rebounded. The title's "from destruction to creation" is that flip.
Hiring Lab stresses this is preliminary evidence, that factors beyond AI complementarities are also at play, and that AI mentions in job titles are spreading from software development to other white-collar occupations. This section draws on a single third-party source with a different lens from the Anthropic session data above; the two are placed side by side as mutual corroboration, not merged into one inference.
The study frames the overall picture as agentic coding amplifying some knowledge while substituting for others. The gains come mostly from competence, not mastery: "proficiency in a domain is enough to use the tool almost as effectively as those with deep mastery."
Stated limits: cannot measure real-world outcomes (whether code is actually used); excludes non-interactive usage (a substantial share); all classifications rely on a model reading the transcript (appendix shows alignment with independent telemetry).
anthropic.com · 2026-06-16 · Economic Research (Zoe Hitzig, Maxim Massenkoff, Eva Lyubich, Ryan Heller, Peter McCrory). All claims, numbers, definitions, and quotes in sections 01–06 and 08 come from this source. All figures are Anthropic's own and cannot be independently verified. Classifiers use Claude Sonnet 4.6; data excludes third-party IDEs, SDKs, and non-interactive claude -p usage. Figure 5's "Beginner" and "Advanced" rows are interpolated from the reported +9%/+13% per-level regression coefficients (the study reports specific numbers for novice, intermediate+, and expert only).
hiringlab.org (Indeed Hiring Lab) · 2026-07-08 · Guillermo Gallacher. All numbers in section 07 (software +15%, overall −7%, senior 71%, AI-title 37%, 27.5% below pre-pandemic) come from this source. This is third-party analysis with a different lens from the Anthropic session data (labor-market demand vs. session-level success); the two are cited side by side, not merged into one inference. The author labels the findings preliminary and stresses that correlation does not imply causation.