Welcome back! Four Google Gemini 4 agents just beat Minecraft's Ender Dragon without a single human stepping in. The agents divided up tasks, shared advice, recovered each other's belongings, and eventually won after 27 hours, 605 deaths, and nearly 2,800 programs written. The whole adventure cost roughly $810 at Gemini's list price. So yes, AI can now beat Minecraft on its own, as long as it has plenty of spare time and money.
Did Microsoft Just Figure Out the Right Way to Add AI to a Product?
Now THIS is how you capitalize on a huge trend and actually listen to what your core users want.
Over the past few months, people have been showing off some genuinely wild AI-powered game mashups. Minecraft characters dropped into Elden Ring, skateboarding through Call of Duty maps, entirely new experiences stitched together from existing games. The trick is using AI coding tools to decompile and recompile games—reverse-engineering their code, remixing it, and shipping new builds on platforms the original developers never supported.
Microsoft saw this trend blowing up and apparently thought… why aren’t WE doing this?
So last week Microsoft announced Xbox Labs, a new experimental division led by CTO Scott Van Vliet (who previously led AI infrastructure at Azure OpenAI). The group’s mandate: AI-powered game mashups, bringing older games to new platforms (browser, Apple Silicon), and making it easier for players themselves to become creators.
A few things that make this more interesting than it looks on the surface:
Microsoft never actually says “AI” in the announcement, even though every single use case is AI-driven. All the PR language is “emerging technologies.”
It’s a direct response to community momentum. Fans decompiling games with AI tooling built the energy. Xbox Labs is explicitly chasing it.
My POV: I think this is a genuinely smart move, and more AI companies should steal the playbook.
Instead of trying to force some random new AI feature into Xbox because the executives are AI-pilled, Microsoft paid attention to what people were already excited about doing with AI and figured out how to build on it with sanctioned tools, legal paths, and a proper distribution channel. That’s a completely different muscle than “we built this cool model, now what should we do with it?”
The trend was already there. Microsoft just decided to run toward it instead of away from it (or in some cases, instead of suing it). I’d love to see more companies follow this exact instinct.
— Matt
Anthropic Pledges 5% of Its Shares to Public Good
Via Anthropic
Anthropic announced a new Public Benefit Fund that would receive 5% of the company's shares, with the potential to grow as Anthropic becomes more valuable.
How it works: The fund will operate as a wholly owned LLC rather than a traditional nonprofit, giving Anthropic flexibility to support different initiatives. The company could make large donations to nonprofits, direct financial distributions to specific populations, and programs addressing AI's impact on employment. Anthropic has already committed $200 million to studying and addressing labor market disruption.
A growing commitment: Subject to shareholder approval, the initial 5% stake could increase by another 2 percentage points for every 50% rise in Anthropic's valuation, at the board's discretion. Combined with existing employee and founder pledges, Anthropic says roughly 20% of its shares are now dedicated to public good. However, the fund's governance, valuation milestones, and decision-making process have yet to be fully disclosed.
The bigger picture: The announcement comes as AI labs face growing scrutiny over safety, employment, and whether the technology's economic benefits will reach ordinary people. Anthropic is making a substantial financial bet that its success should translate into broader public gain. But committing the shares is only the first step. How the money is ultimately distributed, and who gets to make those decisions, will determine how meaningful this promise becomes.
Amazon Makes Its AI Chips Easier to Use
Amazon introduced a new software tool called TorchNeuron that allows developers to run PyTorch code without rewriting their programs on Amazon’s custom Trainium AI chips.
What changed: PyTorch is one of the most widely used frameworks for building AI models, but running that code on different hardware has historically required extra engineering work. TorchNeuron removes much of that friction. Developers can move existing PyTorch workloads onto Amazon's chips with a single configuration change, making it significantly easier to experiment with alternatives to Nvidia.
The Nvidia challenge: Nvidia's dominance isn't just about powerful GPUs. Its CUDA software ecosystem has kept developers tied to its hardware for years. Amazon is attacking that advantage by making its own chips compatible with the tools developers already use. TorchNeuron is free and open source, and Amazon says more than 30 early customers across robotics, voice AI, and world models are already using it.
The bigger picture: Compute has become one of the most valuable resources in the AI economy, and every major tech company wants more control over its hardware supply and costs. Making alternative chips easier to adopt could weaken Nvidia's software advantage while giving developers more flexibility.
What If Your AI Kept Working After You Logged Off?
Most AI waits for another prompt—OpenAI’s dots is, well, connecting the dots. Designed to keep moving between conversations, dots is your always-on agent that takes ongoing responsibilities and runs with them.
Set permissions and boundaries, then let dots:
Research, draft documents, and write and test code using its own cloud computer, browser, and connected apps.
Use relevant ChatGPT memory and your feedback to learn your priorities, preferences, and what “good work” looks like to you.
Keep projects moving between conversations, bring back completed work for review, and ask for your input when a decision needs your judgment.
Ready to take something off your plate? Start delegating a duty today.
Produce Music With AI
Via Veena
Veena is an AI-native digital audio workstation with a built-in agent that helps musicians compose, produce, and mix songs directly inside their sessions. It supports audio and MIDI generation, stem separation, real-time pitch correction, and browser-based recording.
How you can use it
Generate melodies, instrumentals, and MIDI tracks
Compose and mix songs with an AI coproducer
Separate vocals and instruments into stems
Convert vocal ideas into instrumental parts
Pricing: Free and paid
Validate Business Ideas With AI
Via Preuve AI
Preuve AI is a market validation tool that helps founders test business ideas before building. It scans 60+ live sources in 60 seconds and generates a detailed report covering competitor analysis, customer demand, market size, risks, and potential pivots.
How you can use it
Validate startup ideas before investing time or money
Analyze competitors and customer demand signals
Estimate TAM, SAM, and SOM market sizes
Identify business risks and potential pivots
Pricing: Paid
Jobs, announcements, and big ideas
xAI's Grok Bot ships IncidentWatch and IncidentSweep, automating incident detection across customer support queues.
OpenAI acquires Orchestra Bio to speed up AI-driven drug discovery and development.
Anthropic's Claude designs 17 protein binders against scorpion neurotoxins in a 64-hour AI campaign.
Google launches a Gemini Business Hub with financial insights and brand tools for small businesses.
Perplexity brings its Computer agent fully local on Apple Silicon Macs for Pro users.
You gotta try these three Claude prompts! Trust me… Let’s get right into it ⤵️
That’s a wrap! See you this Friday for more.
