Harvard launches a $699 bootcamp taught by AI avatars of its faculty: Harvard started a $699 online startup bootcamp whose instructors are AI avatars (clones) of its professors. It’s a sign that AI-avatar instruction is going mainstream even at elite universities. TechCrunch
DeepMind-alumni startup Inherent says its AI ’teammate’ beats Anthropic and OpenAI at replicating research: Inherent claims its AI teammate outperformed Anthropic and OpenAI models on research-replication tasks — a newcomer aimed at automating the research loop. TechCrunch
OpenAI urges California to strengthen its AI safety bill: OpenAI issued a policy stance urging the California legislature to strengthen its AI safety law — a notable leaning into tighter frontier-model regulation. TechCrunch
Agent interoperability standards are becoming the layer where competition happens: The Model Context Protocol (MCP) team released a new roadmap charting the next phase of the standard for sharing tools and context across agents. Competition in the agent ecosystem is shifting from individual models to the surrounding protocol and infrastructure layer. Model Context Protocol
The ‘what did AI actually return’ measurement gap is widening: A survey finds 90% of executives say AI hasn’t boosted productivity — even as some keep cutting jobs. Measuring and proving AI ROI is becoming the binding constraint on the next buying cycle. Fortune
Build middleware, evaluation, and integration tooling around MCP: A firmer MCP roadmap makes the agent-interop layer more investable — connection, management, and evaluation tooling is a realistic wedge for new builders. Model Context Protocol
Instrumentation and audit tools for AI productivity: With 90% of executives unable to attribute productivity gains to AI, tooling that measures whether AI is actually delivering value is a clear gap. Fortune
AI-avatar training and onboarding products: Harvard shows AI-avatar instruction can be packaged as a product; enterprise onboarding, staff training, and courseware are adjacent markets worth experimenting in. TechCrunch
Frontier labs still won’t say how they’d contain a rogue model: Despite pressure, frontier AI labs haven’t detailed practical containment plans for a misbehaving or runaway model — the governance gap persists. TechCrunch
What actually lands in California’s AI safety law: OpenAI’s support for strengthening the bill is notable, but the final text and how it reshapes the regulatory market remain open questions. TechCrunch