OpenAI reports growing internal use of coding agents alongside increased coding and experimentation. Complex tasks still need human guidance, while people set priorities and make research and deployment decisions. These are preliminary institutional observations, not independent causal evidence; activity measures do not directly establish scientific progress.
Anthropic models how AI capabilities, adoption, autonomy and job-switching frictions could shape US growth, wages and employment through 2030. Across three illustrative scenarios, output rises while knowledge workers may face wage and employment pressure, and capital captures a larger share of income. These are conditional scenarios, not forecasts or probability estimates. The framework omits policy responses and business cycles, among other forces, and cannot be directly generalized to other countries.
Analysis of roughly 400,000 Claude Code sessions finds people usually plan while AI executes. Greater domain expertise is associated with more work per instruction, while users across occupations achieve similar coding-task success rates. Transcript classifications do not establish whether resulting artifacts are deployed or economically valuable.
In a randomized study of 52 developers learning a Python library, immediate quiz scores averaged 50% with AI versus 67% without it; completion-time differences were not statistically significant. Concept-focused interaction was associated with better comprehension, without establishing causality between usage styles. The small, short-term study cannot determine long-term skill development.
DeepMind reports AlphaEvolve applications in sequencing, grid optimization and computing infrastructure. Candidate algorithms are iteratively evaluated and selected, illustrating how testable objectives can guide automated search. Reported gains are institutional disclosures and should not be treated as independently established effects across all settings.
AlphaGenome combines long DNA contexts with predictions across regulatory modalities, matching or exceeding compared external models on 25 of 26 variant-effect evaluations. It can help prioritize and interpret variants for investigation. Molecular predictions still require experimental validation and do not independently establish clinical diagnoses or treatment decisions.
A preregistered experiment with 791 P&G professionals found that individuals using AI matched teams without it on product-innovation tasks and produced more balanced technical and commercial proposals. The journal version distinguishes improved idea generation from the continuing value of human selection. It does not establish replacement of durable team relationships.
Danish surveys linked to administrative records show task reorganization around content generation, AI oversight and integration. Difference-in-differences estimates find no average earnings or recorded-hours effects larger than 2% within two years of ChatGPT’s launch. These early, context-specific results do not imply that future labor-market impacts will remain small.
The report describes AI use through task complexity, skills, purpose, autonomy and success, incorporating success rates into efficiency estimates. Tasks within one occupation can differ substantially, so usage counts alone cannot measure automation. Many metrics rely on model assessment of platform conversations rather than causal measurement of economy-wide productivity.
DeepMind presents Gemini Robotics 2, combining visual-language understanding with whole-body control, dexterity and coordination. Demonstrations extend AI assistance into physical environments. Evidence mainly comes from developer demonstrations and evaluations; it does not establish universal reliability, safety or commercial returns in uncontrolled workplaces.