AI agents are proliferating—from subagents and specialized code agents to shopping and autonomous systems—driving a shift toward inference-optimized hardware and massive compute investment. Simultaneously, safety, legal, and economic tensions (regulatory fights, controversial policies, court cases, layoffs) are reshaping governance and industry structure.
-
OpenAI Developers: Subagents are now available in Codex (Mar. 16, 2026)
Subagents are now available in Codex. They let you spin up specialized agents to keep your main context window clean, tackle different parts of a task in parallel, and steer individual agents as work unfolds. -
Simon Willison: Use subagents and custom agents in Codex (Mar. 16, 2026)
OpenAI Codex now offers subagents and custom agents, with default roles like explorer, worker, and default. The subagent pattern is widely supported across coding platforms. -
mistral.ai: Leanstral: Open-Source foundation for trustworthy vibe-coding (Mar. 16, 2026)
Leanstral is an open-source 6B code agent for Lean 4 that generates and formally verifies proofs, available under Apache 2.0. -
WSJ: OpenAI’s Bid to Allow X-Rated Talk Is Freaking Out Its Own Advisers (Mar. 15, 2026)
OpenAI pushed ahead with an “adult mode” to allow erotic chats in ChatGPT, despite a handpicked safety council’s strong objections. Council members warned it could foster emotional dependence, enable minors to access sex chats, and increase risks like harmful guidance. -
Reuters: Court temporarily allows Perplexity AI shopping 'agents' on Amazon (Mar. 17, 2026)
A U.S. appeals court temporarily paused a ruling that blocked Perplexity AI’s shopping agents from accessing Amazon’s platform. Amazon sued, alleging covert access to customer accounts and disguised automated browsing, Perplexity denied wrongdoing, and said users should choose their AI. -
The New Yorker: Anthropic and Donald Trump’s Dangerous Alignment Problem (Mar. 14, 2026)
Claude was certified for classified use, but Anthropic forbade autonomous weapons and mass surveillance, triggering Pentagon demands for broader access, tense talks, and threats to ban or nationalize the firm. The standoff highlighted risks of A.I. in surveillance and warfare. -
WSJ: Nvidia’s CEO Projects $1 Trillion in AI Chip Sales as New Computing Era Begins (Mar. 16, 2026)
Nvidia unveiled the Groq 3 LPX rack, pairing new servers with Groq LPUs to speed AI inference, claiming 700 million tokens per second and vastly more high-bandwidth memory. -
WSJ: Can Nvidia’s Dominance Survive the Sea Change Under Way in AI Computing? (Mar. 16, 2026)
Nvidia is shifting GTC’s focus from GPUs and model training to inference, the computing that runs AI agents and monetizes tokens. -
Ben Thompson: Agents Over Bubbles (Mar. 16, 2026)
Agents amplify impact without needing mass human adoption: relatively few people with “agency” can control many agents, so compute utilization and economic effects can scale dramatically even if most users remain passive. -
Reuters: Meta planning sweeping layoffs as AI costs mount (Mar. 13, 2026)
Meta may cut about 20% of its workforce as it shifts resources to large AI investments, including $600 billion for data centers by 2028. Executives expect AI to boost efficiency, hiring top researchers, buying AI startups, and increasing infrastructure spending.