Try the free and instant web host for agents

Good morning. It's Monday, March 23, and we're covering OpenAI’s race toward autonomous AI researchers, Trump’s push for national AI rules, and more.

Tesla Plans TERAFAB Chip Factory: Elon Musk's "TERAFAB" is a massive, multi-billion dollar chip fabrication project located in Austin, Texas, designed to be a joint venture between Tesla and SpaceX for manufacturing in-house AI chips.

Trump Pushes National AI Rules: The administration proposed a federal AI framework preempting state law, aiming for uniform standards, IP protections, and faster innovation amid rising global competition.

AI Bots Set to Dominate Web: Cloudflare CEO Matthew Prince said AI-driven bot traffic could surpass human activity by 2027, increasing infrastructure strain as agents query thousands of websites per task.

AI Tokens Enter Pay Packages: Tech firms now offer engineers sizable AI compute budgets alongside salary, equity, and bonuses, with some packages adding $100,000+ in tokens.

NVIDIA CEO Pushes Heavy AI Usage: Jensen Huang said engineers earning $500,000 should use at least $250,000 in AI compute tokens, signaling growing expectations that developers will rely heavily on AI to boost productivity.

Tech Executives Charged in Chip Scheme: U.S. prosecutors charged three people tied to Super Micro with helping smuggle at least $2.5 billion of U.S. AI technology to China, according to the Justice Department.

AI Revives Val Kilmer Role: In “As Deep as the Grave,” filmmakers used generative AI—with family consent—to recreate Val Kilmer’s performance after illness prevented filming, highlighting ethical debates over digital actors.

here.now is web hosting for agents:

Perfect for anything you create with your agent and need a URL for: documents, websites, dashboards, tools, school work, visualizations, presentations, prototypes, games, articles, galleries, portfolios, media files, etc.

Works inside your favorite agents like Claude Code, Codex, Cursor and OpenClaw.

Free, no sign-up required.

Get setup instructions for your agent at here.now

OpenAI is pivoting its long-term research strategy toward building a fully autonomous “AI researcher,” according to chief scientist Jakub Pachocki in an interview published March 20, 2026. The company aims to debut an “AI research intern” capable of handling multi-day tasks by Sept. 2026, with an advanced multi-agent system planned for 2028.

The effort builds on recent advances in reasoning models and agent-based tools like Codex, which OpenAI says already boosts productivity across its engineering teams. If successful, such systems could accelerate breakthroughs in fields like math, biology, and policy—while raising unresolved safety and governance concerns. Researchers outside OpenAI caution that chaining complex tasks remains error-prone, underscoring technical hurdles ahead. → Read the full article here.

Tech workers are increasingly competing to maximize their use of AI tools, with some racking up massive token consumption—and costs—as companies reward heavy usage. At firms like OpenAI, Meta, and Anthropic, internal leaderboards and performance reviews now track AI usage, with one Anthropic user reportedly generating a $150,000 monthly bill.

The rise of autonomous coding agents that run continuously has fueled the trend, enabling billions of tokens to be processed weekly. While companies see AI usage as a proxy for productivity, some employees warn it risks becoming costly “productivity theater” with unclear output quality. → Read the full article here.

OpenAI is refocusing its strategy on coding tools and enterprise customers after internal concerns it has spread itself too thin, according to a March 18, 2026 report. Applications chief Fidji Simo told staff the company will prioritize business adoption as rivals like Anthropic gain traction with developer-focused products.

OpenAI says more than 1 million businesses use its tools, with its Codex coding assistant at over 2 million weekly active users, and is exploring partnerships with private equity firms to expand distribution. The shift comes as some newer products, including its video generator Sora, reportedly see declining usage. The move signals a return to core revenue drivers as competition intensifies and enterprise AI spending accelerates. → Read the full article here.

Chinese AI startup MiniMax has released M2.7, a proprietary reasoning model that can autonomously handle 30–50% of its own reinforcement learning (RL) research workflow, marking a step toward self-improving AI systems. The model uses agent-based loops to debug, optimize, and refine its own training processes, achieving competitive benchmark performance with leading models from Google and Anthropic while maintaining significantly lower costs.

M2.7 also shows improvements in software engineering, document processing, and hallucination reduction compared to its predecessor. Why it matters: the release signals a shift toward recursive AI development—where models increasingly contribute to their own advancement—while highlighting a broader industry move by Chinese firms toward proprietary systems. → Read the full article here.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team