Are AI Builders the Final Abstraction Layer? (fortune.com)

Anthropic has unveiled Claude Sonnet 4.5, pitching it as a step-change in autonomous work and coding: the model reportedly ran autonomously for 30 continuous hours—building an entire software application with minimal oversight—compared with just seven hours for its Opus 4 predecessor. Anthropic says Sonnet 4.5 outperforms earlier Claude variants on key benchmarks (including state-of-the-art results on SWE-Bench Verified) and on practical business tasks like financial research, modeling and forecasting. The company highlights improvements in instruction-following, identifying code improvements, and producing more production-ready code; Sonnet 4.5 also bests Opus 4 on customer-focused metrics and coding benchmarks. The release matters because it signals a move from chat-oriented assistants toward AI builders that can autonomously complete long, complex workflows—particularly in software engineering. Usage data reinforce this shift: 36% of global Claude.ai activity is math and coding, 77% of API prompts request the model perform tasks (not just advise), 44% of API use is coding, and 5% is AI development/evaluation. If models like Sonnet 4.5 reliably sustain long autonomous runs, businesses can automate high-cost, time-intensive workflows, reduce oversight, accelerate delivery, and potentially shrink headcount—raising productivity gains alongside operational and workforce implications for the AI/ML ecosystem.