Microsoft's $13 Billion AI Partnership Could Collapse This Month
4IR - Daily AI News
Welcome back to 4IR. Here's today's lineup:
Microsoft racing to keep OpenAI access after AGI milestone as partnership turns competitive
Trump's AI Action Plan bans "woke" systems from federal contracts while gutting Biden-era oversight
Study finds AI tools make experienced developers 19% slower despite productivity hype
Tools & Trending
🔥 TOP STORY: Microsoft scrambles to solve OpenAI's "AGI doomsday clause"
The story: Microsoft is in urgent talks to fix a major problem buried in its OpenAI contract. There's a clause that cuts off Microsoft's access to OpenAI's technology once artificial general intelligence is achieved—something that seemed far-off when they signed the deal but now feels uncomfortably close. Talks could wrap up in weeks, with Microsoft pushing for a 30%+ stake in whatever OpenAI becomes.
What we know:
Microsoft loses all IP rights once OpenAI hits AGI or 2030, whichever comes first
Microsoft just blocked OpenAI's $3 billion Windsurf acquisition over IP concerns, showing the relationship is getting tense
OpenAI has been building backup plans, signing nearly $16 billion in deals with Oracle, Google Cloud, and CoreWeave
Why it matters: Microsoft built everything around OpenAI's models—Azure OpenAI Service, Copilot across Windows and Office, GitHub integration. What started as a partnership has become a rivalry as both companies chase the same enterprise customers. Microsoft is already preparing Copilot for GPT-5's expected August launch with a new "Smart" mode, but they need to secure access first.
The timing reveals just how much pressure both companies are under. With Elon Musk's lawsuit challenging OpenAI's for-profit transition and regulators circling, they need to get this done. Sam Altman's recent tweet about their partnership's "next phase" being better than anyone expects now sounds more like spin than confidence.
🏛️ POLICY SHIFT: Trump administration launches sweeping AI deregulation push
The story: Trump rolled out his "America's AI Action Plan" last week, which frames removing regulations as boosting innovation while dismantling most of Biden's AI oversight. The key move is requiring federal contractors to prove their AI systems are "free of ideological bias" to keep government contracts.
What we know:
Companies must certify their AI doesn't include things like "critical race theory" or "unconscious bias" to stay eligible for federal work
The plan fast-tracks permits for data centers and chip facilities while removing climate requirements from CHIPS Act funding
Federal agencies can now cut AI funding to states with "burdensome" AI rules
Why it matters: This is a complete reversal from Biden's approach, which focused on preventing AI discrimination. Critics say it's just corporate deregulation dressed up as innovation policy. The problem is nobody can actually define what "ideological bias" means in AI systems. White House officials couldn't explain how to measure it, leaving companies to guess what compliance looks like. Meanwhile, China just launched its own AI plan positioning itself as the more inclusive alternative.
The bigger issue: how do you build AI that works for everyone when you can't even agree on basic facts?
🎯 REALITY CHECK: Careful study punctures AI productivity hype
What happened: METR researchers did something rare, they actually measured AI's productivity impact instead of relying on surveys and hype. They found experienced developers took 19% longer to finish tasks when using AI tools like Cursor Pro and Claude. This wasn't a quick study either, they tracked 16 seasoned developers across 246 real tasks on major open-source projects.
The disconnect: Developers predicted AI would make them 24% faster. After getting slower, they still thought they were 20% faster. The gap between what people think is happening and what's actually happening is huge.
What went wrong: Developers spent more time babysitting AI suggestions than the AI saved them. One developer wasted an hour trying to get AI to solve a problem before giving up and coding it manually. The AI was typically "directionally correct but not exactly what's needed", which might be the perfect description of AI in 2025.
Why this matters: Software runs everything, and we're rapidly filling codebases with AI-generated code that fewer humans actually understand. As one expert put it, companies are about to learn what happens when their code gets "infiltrated with AI-generated code at scale", and it probably won't be pretty when something breaks.
🛠️ HOW-TO: Set up AI meeting automation that actually works
What you'll build: A system that turns your meetings from time sinks into action generators through automated capture, analysis, and follow-through.
Steps:
Get Otter.ai or Fireflies.ai running in every important meeting, just set it up once and let it capture everything
After each meeting, feed the transcript to ChatGPT or Claude with this prompt: "Pull out action items with owners and deadlines, identify any risks that need follow-up, suggest what meetings should happen next, and rate how effective this meeting was with your reasoning"
Connect the output to your project management tool so tasks get created automatically and people get notified
Set up rules like "if the meeting mentions budget concerns, alert the finance team" so nothing falls through the cracks
The results: Companies report cutting meeting admin time by 90% and following through on 75% more action items. One team saw their projects move 25% faster just because stuff stopped getting lost.
Pro tip: The key is automating the whole chain so humans don't have to touch it. Start with your most important meetings to prove it works, then expand.
⚡ QUICK HITS
New Tools
Triple Whale Moby — E-commerce AI that actually optimizes your spending automatically instead of just reporting on it
Google Virtual Try-On — Try on clothes from billions of items across Search and Shopping, with custom price alerts
Constructor AI Shopping Agent — Natural language shopping that doesn't suck, now available on AWS Marketplace
PCI Pal Fraud Management Suite — AI fraud detection that stops bad actors before they get your money
Other News
AWS Summit showed off new agentic AI features for automating complex business processes
Google's "Big Sleep" AI hunts down sketchy websites before they become problems
Scientists used AI to design paint that keeps buildings 30% cooler by reflecting more sunlight
EU delayed its AI guidance until late 2025, giving companies more time to figure out compliance