The pope wrote 42,000 words about AI.

A venture capitalist made one phone call and killed a White House executive order before it could be signed.

And Meta surveilled thousands of workers for months, learned everything they do, built AI to replace them, and then fired eight thousand of them.

If you missed it, this was one of the most consequential weeks in AI in recent memory. Not because of the hype, but because of what quietly changed. Google dropped over 100 updates. OpenAI's Codex got capabilities that will genuinely change how teams work. And the big AI labs stopped just selling models. Now they're embedding their engineers inside your competitors' companies to run AI strategy for them.

We tackled it all today on Everyday AI, and here's what you actually need to know.

Google used its I/O developer conference this past week to announce more than 100 AI updates, led by three things worth actually paying attention to.

First, Gemini 3.5 Flash. Google CEO Sundar Pichai says it beats the roughly four-month-old Gemini 3.1 Pro on nearly every benchmark, runs significantly faster than other frontier models, and is optimized for longer autonomous task sessions.

Second, Gemini Omni. A multimodal model rebuilt from the ground up to eventually handle text, video, drawings, and more. Right now the output is video only, and critics are already comparing it unfavorably to Chinese competitors. The bigger vision is still ahead.

Third, Gemini Spark. Google's always-on personal agent that runs in Google Cloud 24/7, even when your computer is closed. Currently in beta for Ultra subscribers in the US only.

Google also knocked the Ultra plan price from $250 down to $200, matching the $200 Claude Max and ChatGPT Pro plans.

What it means: Google had a massive week on paper. But the classic Google problem is still very much alive. Gemini Spark is rolling out to a fraction of users. Gemini Omni is limited to video at launch. Watch what actually ships broadly, not just what gets announced on stage.

OpenAI shipped several major Codex updates this past week, and if there's one story that matters most for how knowledge workers actually operate, this is it.

Appshots is now live on macOS. Press both Command keys and whatever app window you're working in gets sent directly to Codex, screenshot and all available text included. No more manual copying or describing what's on your screen.

Goal Mode is out of experimental status and fully available. Hand Codex an open-ended objective and it works toward that goal autonomously for hours or even days. This is the first true goal mode made widely available for everyday knowledge workers from a major AI company.

Yuuuuup, there's more. Codex can now operate desktop apps after a Mac is locked, including remotely from your phone. And plugin sharing lets teams share reusable plugin bundles across their entire organization.

What it means: Goal Mode going fully available is a legitimately big shift. Assigning a multi-day autonomous objective to an AI agent and walking away is exactly the capability that starts changing how organizations actually operate. If you're on the $20 plan, monitor it closely. It can burn through credits fast.

A federal jury in Oakland made short work of one of tech's most high-profile legal battles last week, unanimously ruling that Elon Musk waited too long to sue Sam Altman, Greg Brockman, and OpenAI.

The jury deliberated for less than two hours.

OpenAI's cofounders discussed a for-profit shift as early as 2017. The for-profit arm was created in 2019. Musk didn't file suit until 2024, blowing past a three-year deadline on the charitable trust claim and a two-year deadline on the unlawful enrichment claim.

The court never touched the actual substance of the case. Musk's team says they'll appeal, arguing it was decided on a calendar technicality.

Sheesh. Honest read: this was theater from the jump. Both companies are heading toward IPOs, and this saga has always been more about leverage and narrative control than legal principle.

What it means: The core claim, that Altman and Brockman abandoned a charitable mission for personal profit, was never actually tested in court. Anyone trying to understand what this was really about should remember both men have companies worth hundreds of billions of dollars and every incentive to control how this story gets told.

The Trump administration had a major AI executive order ready to sign last Friday. It would have created federal oversight for frontier AI models, with the NSA having final say on which systems qualified for review.

Then David Sacks, venture capitalist and Trump adviser, reportedly called the president directly on Thursday morning without telling most of his own staff and derailed the entire thing.

Sacks had previously reviewed the draft and was believed to be supportive. He changed his position late Wednesday night, arguing a voluntary review process could eventually become mandatory and hand what he called "AI doomers" a regulatory foothold.

The major AI company CEOs had already flown to Washington for the signing ceremony. Word came: it was off.

What it means: One phone call, made outside the normal policy process, stopped one of the most significant AI governance moves in years. Let's be honest: the 90-day model review window was probably too long for how fast labs actually move. But the way this collapsed, with a single adviser calling the president directly and blindsiding his own staff, is a preview of how AI governance keeps getting derailed by the people with the most financial interest in less of it.

According to Reuters, Meta cut about eight thousand US employees last week, roughly 10% of its workforce, as the company redirects spending toward AI infrastructure. Another 6,000 open roles were canceled, and roughly 7,000 current employees were moved into AI-focused work with many non-AI titles collapsed into a broader "AI builder" category.

But the layoffs aren't even the part grabbing headlines.

It's a program called the Model Capability Initiative. An internal monitoring system that logs employees' mouse movements, clicks, and keystrokes on company laptops to collect training data for AI systems learning how employees complete routine digital tasks. No opt-out.

More than 1,500 employees signed a petition against it. Workers started posting anti-surveillance flyers around Meta offices.

The timeline is brutal: months of monitoring, everything learned, and then the jobs disappear.

What it means: Welp. This is going to be the blueprint studied for the next decade. Cut humans, automate their roles, use their own behavioral data to do it. Meta is betting the long-term upside outweighs the PR damage. If the stock holds, other large enterprises will be drawing their own conclusions very quickly.

Reports confirmed this past week that Microsoft and EY have expanded their partnership with a commitment of more than $1 billion over five years to help enterprises scale AI. The arrangement puts Microsoft forward deployed engineers, or FDEs, directly inside EY's client teams to speed up implementation.

Microsoft is technically the last major player to make this pivot.

In a six-week stretch, OpenAI, Anthropic, and Google all made the same move. OpenAI launched DeployCo on May 11, a $4 billion deployment operation with 19 partners including TPG, Bain Capital, and McKinsey, embedding FDEs directly inside client companies. Anthropic followed with a $1.5 billion joint venture with Blackstone and Goldman Sachs targeting mid-market companies. Google committed $750 million to 120,000 partners including Accenture and Deloitte to build Gemini agents.

What it means: The AI model business is quietly becoming the AI implementation business. Embedding your engineers inside the companies that pay the most, and building structural dependency on your platform, is a far stickier long game than selling model access. Your competitors who move first on this get a significant head start.

Pope Leo XIV issued his first encyclical last week, a more than 42,000-word document called "Magnifica Humanitas," warning that AI could reshape society in dangerous ways if not guided by human values.

He signed it 135 years after Pope Leo XIII's "Rerum Novarum," the foundational Catholic teaching on industrial-era labor and justice. The parallel is entirely intentional.

Leo says AI is useful but not neutral, reflecting the priorities of whoever builds, funds, and deploys it. He warns AI could weaken human judgment, imitate care without genuine relationship, accelerate disinformation, concentrate power in a handful of companies, and lower the barrier to lethal military decisions.

Experts say the document carries weight well beyond Catholic circles as governments debate AI rules for hiring, credit, public services, and safety.

What it means: Nah, this isn't just a religious document. A 42,000-word moral framework from the head of 1.4 billion people carries real cultural and policy weight. The concentration of AI power, the disinformation problem, the false intimacy of chatbots — these tensions are playing out in real time, at exactly the moment AI systems are becoming more persuasive than anything that's existed before.