AI in the workplace** Task automation and augmentation Economic impact of AI

“Which Economic Tasks are Performed with AI? Evidence from Millions of Claude Conversations”

Posted on October 01, 2025 at 09:23 PM

1. Research topic and objective

Topic The study examines how people are actually using AI—specifically through Anthropic’s Claude model—in real economic tasks, mapped to occupations and work functions. It seeks to move beyond speculation and expert forecasts, instead providing empirical evidence of AI adoption in real-world usage.

Objective

  • To develop a method to map anonymized AI interactions to specific tasks and occupations (via U.S. Department of Labor’s O*NET database)
  • To quantify how many tasks and which occupations are affected by AI
  • To distinguish augmentation (AI assisting a human) versus automation (AI autonomously doing a task)
  • To provide a baseline / “index” that can be tracked over time to understand AI’s evolving role in the labor market

2. Key findings and conclusions

Here are the main outcomes and insights:

  • Concentration in coding + writing tasks AI usage is heavily concentrated in software development (coding) and writing tasks. Together, these two categories account for nearly half of total usage observed. (arXiv)

  • Broad but uneven usage across occupations While AI use is concentrated, it is not limited to only a few fields: about 36% of occupations use AI for at least 25% of their associated tasks. (arXiv)

  • Augmentation > Automation Of all the AI‐human interactions studied:

    57% of usage suggests augmentation (i.e. the AI helps, iterates, or is used interactively with human oversight) 43% suggests automation (i.e. the AI performs a task with minimal human involvement) (arXiv)

  • Variation by task and occupation Some tasks show more automation (e.g. translation) and others show more human-AI co-writing or iteration (e.g. editing or creative writing). (Anthropic)

  • New “extended thinking” mode usage In later updates (particularly after the release of Claude 3.7), the model’s “extended thinking” mode (which allows the model to reason more deeply or hold longer chains of thought) is used more in technical, scientific, and creative tasks (e.g. by software developers, multimedia artists). (Anthropic)

  • Limitations acknowledged The authors are careful to note limitations:

    • The data is only from interactions with Claude (not all AI systems).
    • Mapping a conversation to a task/occupation is imperfect.
    • It is not always clear whether outputs are used, modified, or discarded by users.
    • The dataset is U.S.-focused and may not generalize globally. (arXiv)

  • Conclusion The authors argue that AI is currently acting more as a complement to human labor than as a wholesale replacement. Their framework also provides a baseline for monitoring AI’s economic impact over time.

3. Critical data and facts

Here are some of the important data points and facts:

Metric / Fact Value / Description
Number of conversations analyzed Over 4 million anonymized Claude conversations (arXiv)
Share of occupations with ≥25% tasks using AI ~ 36% of occupations (arXiv)
Augmentation vs Automation ratio 57% augmentation, 43% automation (arXiv)
Proportion of usage in coding + writing Nearly half of total usage (arXiv)
Use of “extended thinking” mode More frequent in technical, creative tasks (e.g. software developers, multimedia roles) (Anthropic)

Additionally, the project has published associated datasets (for tasks, occupation mappings, wage data, etc.) under open licenses. (Hugging Face)

4. Potential applications or implications

Based on the findings, here are several implications and possible applications:

  1. Policy & workforce planning Governments and institutions can use this empirical view to guide training programs, education updates, and labor policies. For instance, they might prioritize helping workers in tasks or fields likely to be impacted by automation.

  2. Tracking AI’s economic impact over time Because Anthropic intends this “index” to be updated periodically, researchers can observe trends: which occupations adopt AI most, shift in augmentation vs automation, etc.

  3. Business strategy & productivity tools Companies can better understand which parts of workflows are most amenable to AI assistance (e.g. coding, writing, editing) and build tooling accordingly.

  4. Bridging inequality risks The concentration of AI effects in high-skilled tasks suggests a risk of widening inequality between workers and regions with high knowledge-work density vs others. Policy interventions may be needed to ensure broader access. (Some media coverage already flags this concern.) (Axios)

  5. Benchmarking for future studies Academics and industry researchers can use the data and methodology as a baseline or comparison for studies across models, geographies, or over time.

  6. Caution in generalization As the authors note, their results apply to Claude users (Free & Pro) and U.S.-based occupational mapping. Other platforms, sectors, or countries may show different patterns.

FEATURED TAGS
computer program javascript nvm node.js Pipenv Python 美食 AI artifical intelligence Machine learning data science digital optimiser user profile Cooking cycling green railway feature spot 景点 work technology F1 中秋节 dog setting sun sql photograph Alexandra canal flowers bee greenway corridors programming C++ passion fruit sentosa Marina bay sands pigeon squirrel Pandan reservoir rain otter Christmas orchard road PostgreSQL fintech sunset thean hou temple in sungai lembing 海上日出 SQL optimization pieces of memory 回忆 garden festival ta-lib backtrader chatGPT stable diffusion webui draw.io streamlit LLM AI goverance prompt engineering fastapi stock trading artificial-intelligence Tariffs AI coding AI agent FastAPI 人工智能 Tesla AI5 AI6 FSD AI Safety AI governance LLM risk management Vertical AI Insight by LLM LLM evaluation AI safety AI Governance Privacy & Data Protection Compliance Microsoft Scale AI Claude Anthropic 新加坡传统早餐 咖啡 Coffee Singapore traditional coffee breakfast Quantitative Assessment Oracle OpenAI Market Analysis Dot-Com Era AI Era Rise and fall of U.S. High-Tech Companies Technology innovation Sun Microsystems Bell Lab Agentic AI McKinsey report Dot.com era AI era Speech recognition Natural language processing Privacy Google Enterprise AI Nvdia AI cluster COE Singapore Shadow AI AI Goverance & risk Tiny Hopping Robot Robot Materials SCIGEN RL environments Reinforcement learning Continuous learning Google play store AI strategy Model Minimalism Fine-tuning smaller models LLM inference Closed models Open models Privacy trade-off MIT Innovations Federal Reserve Rate Cut Mortgage Interest Rates Credit Card Debt Management Investor Sentiment Enterprise AI adoption AI Innovation AI Agents AI Infrastructure Generative AI Workslop Federal Reserve Government Shutdown AI Fine-Tuning LLMOps Frontier Models Hugging Face Multimodal Models Energy Efficiency AI infrastructure Semiconductors Gold & index inclusion