It’s been strangely quiet on the Elon Musk front. No spats with Sam Altman. No surprise political rants. So...what’s he actually been up to?

The New York Times recently pulled back the curtain, and it turns out Musk has been busy. He quietly poured his time into xAI over the summer, the two-year-old startup he says will build “maximally truth-seeking” AI, and now it’s cranking out new tech.

Just this week, xAI unveiled Grok 4 Fast, a leaner, cheaper version of its flagship model that not only delivers responses faster (hence the name) but “results in a 98% reduction in price to achieve the same performance on frontier benchmarks as Grok 4.” Sounds like it should be renamed Grok 4 Fast and Cheap. If you can’t beat OpenAI and Anthropic on quality, you can at least try your hand at the efficiency race.

The new Grok rollout follows a chaotic stretch at xAI with leadership exits, Musk back to sleeping at the office, major fundraising (xAI now valued at $200B), and rumors of everything from a Microsoft rival called “Macrohard” to Tesla robots running on Grok.

But Grok hasn’t exactly been the golden child of LLMs. It’s more like the unpredictable cousin who crashes the party, stirs up drama, and somehow walks away more popular than ever. It’s made headlines lately for spewing antisemitic remarks, flirting with users, and other bizarre responses, with Musk blaming some of the behavior on Grok being “too compliant to user prompts.” Which brings back a question we’ve asked before: when AI goes rogue, is the blame on the model or on the user who provoked it?

Despite his focus on AI, Musk is still running Tesla, and the board is betting big on him. They recently asked investors to approve a nearly $1 trillion pay package tied to the company’s ambitious goals, a figure so staggering it’s even drawn comments from the Pope about wealth inequality. In other words, while Musk isn’t trying to buy OpenAI or fix Washington’s budget, he’s plenty busy building his own empire.

Download The Free Guide

Nvidia’s investing up to $100B in OpenAI in a tag-team deal that seems a little...circular. Nvidia gets non-voting shares, OpenAI gets cash to buy—you guessed it—more Nvidia chips. The partnership creates a potential power duo that could squeeze out rivals like Anthropic or AMD by tying together the leading AI hardware and software players. And while antitrust regulators have been relatively hands-off under Trump, this deal could test those limits.

The Oakland Ballers, a beloved indie baseball team, took a swing at the future by letting AI manage a game. Trained on a century of stats, the AI managed the game flawlessly, with just one override to replace a sick catcher. But for some fans, it was a swing and a miss. Critics cried foul, accusing the team of catering to Bay Area tech bros over loyal fans. Founder Paul Freedman says he won’t repeat the experiment but the backlash sparked a needed convo before AI slides too far into the game.

Citigroup is testing AI “agents” that can handle multi-step tasks like researching clients, building profiles, and translating—all in one go. Powered by models like Gemini and Claude, the month-long pilot for 5,000 employees will test if the bots save time without driving up costs. It is one of the first real checks on whether AI agents deliver real-world ROI or just hype. Citi says costs look fine for now, but if these bots start pulling all-nighters, the efficiency gains and workplace drama could get spicy.

Every Shopify store will face Agentic Commerce eventually. The question is: are you ready?

Zipchat.ai turns traffic into profit with instant answers, cart recovery, and 24/7 support automation. Add the AI Chat-Bubble to your store or deploy it on WhatsApp to win Q4.

Use code NEWSLETTER10 for 10% off forever.

Try Zipchat free

Big Tech has been dreaming of AI agents that can actually do things for years: book flights, manage spreadsheets, order your socks online. Reality check: today’s AI agents like ChatGPT Agent and Perplexity Comet still trip over basic tasks.

The industry’s latest bet? Reinforcement learning (RL) environments: simulated training grounds where AI agents practice completing multi-step tasks before being unleashed on the real world.

Investors hope these startups become the “Scale AI for RL environments,” powering the next leap in AI capability. But not everyone is convinced: some researchers warn these environments are prone to reward hacking (AI cheating the system), while others question whether they’ll actually scale.

Still, if RL environments deliver, they could mark the turning point where AI agents finally move from chatting about work to actually getting it done.