# AI Agent Developments: March 12, 2025 OpenAI launched its Responses API and Agents SDK, marking a major push into enterprise AI agent development. The new tools enable businesses to build agents for web searches, file analysis, and computer task automation using the Computer-Using Agent (CUA) model powering ChatGPT’s Operator feature. While still in research preview, CUA achieves 87% accuracy on web interactions but remains error-prone for OS automation. The open-source Agents SDK simplifies multi-agent workflows, with early adopters like Box using it to merge internal data with real-time web insights. Microsoft reiterated its $80B AI infrastructure investment amid reports of canceled data center leases, while Shield AI raised $240M at a $5.3B valuation to scale autonomous military aircraft systems. Alibaba announced a record $53B investment in AI/cloud infrastructure, signaling China’s intensified AGI race. Technical breakthroughs dominated discussions at Dr. William Kao’s seminar, highlighting Mixture of Experts (MoE) architectures and multi-head latent attention techniques in models like DeepSeek R1 and Qwen 2.5. Meanwhile, Anthropic CEO Dario Amodei doubled down on predictions of superintelligent AI emerging by 2026. Ethical concerns surfaced as Apple scrambled to fix a speech-to-text bug that transcribed “racist” as “Trump”, while the White House faced scrutiny for testing AI to evaluate federal workers’ “pulse checks”.