Most executives building an AI strategy right now are making decisions based on a landscape that no longer exists.

This week alone: OpenAI's leadership bench got thinner right before its historic $122 billion close, Anthropic shut down one of the most widely used open-source AI agent setups on the planet, and Google made a compelling case that frontier-level AI might not require a monthly subscription anymore.

Are you keeping track here? 

That’s why our ‘AI News That Matters’ Monday show exists: so none of that catches you off guard and you’re making the right moves knowing all the pieces on the board.  

Here's the full breakdown of what you may have missed in AI this week and why it matters.

Multiple reports confirmed this past week that Fidji Simo, OpenAI's CEO of AGI Deployment, is taking several weeks of medical leave to address a relapse of POTS, a chronic neuroimmune condition she's spoken publicly about for years.

Her absence triggered a broader reshuffle.

COO Brad Lightcap transitions to a special projects role centered on a reported $10 billion joint venture with private equity firms to distribute OpenAI's technology, reporting directly to Sam Altman, while President Greg Brockman steps in to lead the product organization in Simo's place.

Chief Strategy Officer Jason Kwon, CFO Sarah Friar, and Chief Revenue Officer Denise Dresser will collectively absorb oversight of business and operations.

Then there's CMO Kate Rouch, who announced she's stepping down entirely to focus on her cancer recovery.

All of it happening simultaneously. All of it right before an IPO.

What it means: Each individual reason is valid and deeply human. But the collective optics are messy for a company preparing to go public. Watch how fast OpenAI fills these gaps. The speed of their response will signal a lot about how ready they actually are for Wall Street.

Anthropic officially announced that Claude Pro and Max subscribers can no longer use their subscriptions to power third-party agentic tools, with OpenClaw — the open-source agent framework that NVIDIA CEO Jensen Huang once called potentially the most important open-source project ever — named explicitly as the first casualty.

The technical reasoning is legitimate: third-party tools bypass Anthropic's prompt caching systems entirely, which means one heavy OpenClaw session burns dramatically more compute than that flat $200 monthly subscription was ever built to handle.

The timing, though? That's a different conversation.

OpenClaw creator Peter Steinberger joined OpenAI back in February, and Anthropic dropped this crackdown just weeks later.

Sheesh.

Heavy users are now staring down cost increases estimated at up to 50 times their previous monthly spend if they stick with Claude through direct API billing.

What it means: Any agentic workflow currently running through a Claude subscription and a third-party tool is operating on borrowed time, because this policy will expand well beyond OpenClaw. Price out API billing now, seriously explore OpenRouter, and start evaluating cheaper open-source alternatives before Anthropic forces the decision for you.

Salesforce unveiled more than 30 new Slackbot capabilities this past week, representing the most significant overhaul of the platform since Salesforce acquired Slack for $27.7 billion back in 2021.

The headline capability is meeting intelligence: Slackbot can now join any video call across Zoom, Google Meet, or other platforms, transcribe the full discussion via desktop audio access, summarize decisions, and log action items directly into Salesforce CRM — automatically, with no human in the loop.

Slackbot also now operates entirely outside of Slack itself through a new desktop agent that reads your screen and works across your other apps without any additional installation.

AI Skills let teams define a workflow once and trigger it on demand across any scenario, while Model Context Protocol integration connects Slackbot to more than 6,000 apps in the Salesforce ecosystem.

The lightweight built-in CRM for small businesses is the sleeper feature most people are sleeping on.

What it means: If your team already lives in Slack and Microsoft Copilot hasn't clicked, this is the moment to stop waiting and start testing. Meeting transcription plus automatic CRM logging is the combination that could actually stick — check your plan tier now, because Business+ and Enterprise+ subscribers get these features first.

Yuuuuup. This is legitimately the story of the week, even though most people completely buried it.

Google launched Gemma 4 this past week, a family of four open-weight AI models all built on the same underlying research as Gemini 3, ranging from a two-billion-parameter edge model that runs completely offline on older smartphones and Raspberry Pi devices all the way up to a 31-billion-parameter dense model that currently sits third on the Arena AI global leaderboard — and not third among open models, but third overall, beating proprietary giants at a fraction of their size.

It outperforms models up to 20 times larger on benchmarks.

All four are multimodal, processing video, images, and text natively, and for the first time in the Gemma family's history, everything ships under the Apache 2.0 license, which removes every legal barrier to commercial use that previously made enterprise teams walk away.

Download it today. Run it offline. Beat what you paid for last year. For free.

What it means: The two biggest objections to local AI — performance and legal clarity — just evaporated in one release. If open-source models are anywhere in your AI strategy conversation, test Gemma 4 this week against the cloud API you're currently paying for and see what the math looks like.

OpenAI officially closed its $122 billion funding round at an $852 billion post-money valuation, making it the largest private fundraise Silicon Valley has ever produced.

SoftBank co-led the round alongside Andreessen Horowitz and D.E. Shaw Ventures, with Amazon committing up to $50 billion, NVIDIA and SoftBank each putting in $30 billion, and roughly $3 billion flowing in through bank channels from individual investors — the first time OpenAI has ever opened that lane.

The business context: ChatGPT now has more than 900 million weekly active users, over 50 million paid subscribers, and the company generates over $2 billion in monthly revenue after posting $13 billion for all of 2025.

Still not profitable. Doesn't expect to be until around 2030.

What it means: Outside the AI bubble, most people just call it ChatGPT — and that brand recognition carries genuine IPO value that competitors can't easily replicate. But five executives shuffling roles the same week this round closes hands Wall Street an easy narrative. Watch the next 90 days closely.

A packaging error in a routine Claude Code release accidentally bundled a debug file into the public npm registry, and that file pointed directly back to Anthropic's own cloud storage, which contained nearly 2,000 files and more than 500,000 lines of fully unobfuscated source code that was never meant to see daylight.

Developers found it within hours.

Mirrors spread across GitHub almost instantly, a clean-room rewrite called claw-code became the fastest-growing repository in GitHub history by hitting 100,000 stars in a single day, and the DMCA takedown notices Anthropic fired back accidentally swept up thousands of completely unrelated repositories before having to be retracted.

Anthropic confirmed no customer data was exposed and called it human error.

Now for the part that gets genuinely complicated: multiple senior Anthropic employees have publicly stated that Claude writes nearly all of its own code, and courts have ruled that AI-generated work cannot be copyrighted, which means outside developers rewriting Claude Code using AI may be operating in a legal gray zone that Anthropic's own public statements helped create.

Chinese clones are already multiplying.

What it means: Leaking proprietary source code of your most popular product category right before an IPO is REALLY bad. 

And if Wall Street analysts start asking hard questions about whether AI-generated software can actually be protected by copyright law, Anthropic is going to have a very uncomfortable conversation before they go public when it comes to dealing what comes after this leak. 

(Hint: expect to see many new coding harnesses like Claude Code that’ll be REALLY good.)