Welcome, humans.

Did y’all see the “world’s-first” (we sure about that?) human vs robot cage match this weekend?

IDK how dude did it, but he took a ~4 foot steel kick to the chest-plate like a champ. Doing the human species proud, my guy!

Here’s what happened in AI today:

• 😼 Amazon blocked Muse; Shopify jumped in.

• 📰 OpenAI and Anthropic neared mutual stress tests.

• 📰 OpenAI proposed recursive self-improvement standards.

• 🍪 Googlebook built Gemini into the laptop.

• 🔧 Grok 4.7 hit 64% on electrical engineering.

😼 Amazon blocked Meta’s Muse, exposing AI agents’ permission problem

Meta built Muse to do the stuff AI agents keep promising to do: open websites, fill forms, book things, and buy things for you.

Then Amazon said, in very Wizardly fashion: you shall not pass!

Did I say wizardly? I meant Miserly… More Scrooge than Gandalf TBH.

Here’s what happened: Amazon blocked Muse from shopping on Amazon.com after saying Meta did not get permission to access the store, the agent did not identify itself while browsing, and its handling of credentials raised security concerns.

• Muse can browse the web and take actions on a user’s behalf, including purchases.

• Meta says Muse runs inside a dedicated secure virtual machine and asks for approval before sensitive actions.

• Amazon says Muse was still an unauthorized third-party agent on its platform, so it cut access.

Here’s the useful distinction: capability isn’t permission. An agent can know exactly how to buy a stroller and still get stopped at the website’s front door.

But Amazon immediately hit the awkward part of blocking an agent: the agent can route around you. Shopify CEO Tobi Lütke jumped in with a deep Muse partnership, putting Shop Pay checkout across Shopify’s store network. signüll thinks this could end with Meta simply paying Amazon for access, creating a new business-to-agent toll.

This is where Ben Thompson’s Aggregation Theory gets interesting. The most powerful layer owns the customer relationship, aggregates demand, and makes suppliers hot-swappable. If Muse knows what you want, remembers your preferences, and controls checkout, Amazon becomes one possible service provider instead of the place you start.

Nicolas Bustamante’s read: smaller services have every incentive to open clean agent access because being routed into beats being routed around. Amazon has enough leverage to resist. Plenty of everyone else may not.

Our take: the next big agent benchmark may be boring old permissioning: can my agent actually use the dang app I need it to use? But the bigger fight is who becomes the aggregator between you and the hot-swappable service providers underneath.

Watch who opens agent access, who charges for it, and who keeps the customer relationship.

FROM OUR PARTNERS

Build AI, Not Infrastructure

AI teams shouldn’t spend their time sizing compute, tuning clusters, chasing permissions, or managing infrastructure. Teradata’s Autonomous Knowledge Platform is designed to remove that friction—helping teams build, scale, and govern AI workloads while the platform handles performance, cost, and infrastructure behind the scenes. 

Your AI team has better things to do than manage infrastructure. 

Getting an AI model to work in a notebook is one thing. Getting it into production—without runaway costs, infrastructure bottlenecks, or endless tuning—is another. 

Teradata’s Autonomous Knowledge Platform uses AI-powered agents to help automate scaling, optimize resources, enforce guardrails, and manage production workloads so developers can stay focused on building AI instead of babysitting infrastructure. 

See what it looks like when the platform does more of the operational heavy lifting. 

See how it works 

🎓 AI Skill of the Day: Don’t /compact your agent just because you’re taking a break

Long coding-agent sessions can feel messy, so /compact looks like housekeeping. Don’t treat it that way.

Kun Chen points out that manual compaction usually triggers a summarization pass over the whole current context. On a huge session, that can itself be expensive, and the compacted session may have to rebuild context without the same cache savings.

Better rule: let the agent harness auto-compact at its tuned threshold. Codex, for example, automatically compacts once its token limit is exceeded. Use manual compaction only when context pressure is actually hurting the task, not before lunch because the context meter looks ugly.

Have a specific skill you want to learn? Request it here.

FROM OUR PARTNERS

You Read About AI Daily. Now be in the Room.

In three days, you will learn more about what actually works than in six months of newsletters. The people building these systems will show you their playbooks, their failures, and their roadmaps. 5,500+ of your sharpest peers will be there.

Code NEURON30 saves 30%. Expires in 24 hours.

Register Now

🍪 Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here!

1. *Build your own AI coworker in just 3 hours (for $0). Join Outskill’s live workshop to learn the ins & outs of the 15 most powerful AI tools right now.

2. Googlebook puts Gemini into the laptop with Magic Pointer, Antigravity agents, and an isolated Linux environment for autonomous work; starts at $899.

3. Cloudflare Python Workers runs FastAPI, Django, Flask, OpenAI, LangChain, and MCP code at the edge without JavaScript glue.

4. Pexo turns a URL, PDF, image, audio file, or rough idea into a video you refine conversationally, including by marking frames to request fixes.

5. Qwen RecreationWorld tests computer-use agents by having them inspect a working app and rebuild it across Ubuntu, macOS, Windows, Android, and the web.

6. slop-grader checks every line against your own writing rules, flags high-confidence violations, and returns structured results an agent can use to rewrite them.

7. P.S: If you don’t get Jev, which we talked about here, watch this interview.

📰 Around the Horn

1. OpenAI and Anthropic reportedly neared a binding agreement to stress-test each other’s commercial models, while Elon Musk separately called for rival labs to test one another before release.

2. OpenAI called for shared technical standards around recursive self-improvement and said fully autonomous RSI should not be pursued until it can be done safely.

3. Google open-sourced AX, a Kubernetes-like orchestrator for running stateful AI agents in isolated workspaces with controlled networking, credentials, tools, and suspend/resume support.

4. Microsoft Research, GSK, and Novartis published RetroChimera, an AI model that improved retrosynthesis predictions across public and proprietary pharmaceutical reaction data.

5. The UN’s independent AI panel published a brief arguing that more capable agents can make misbehavior and concealment harder to detect, based partly on recent OpenAI-Hugging Face incidents.

🔧 Tuesday Tech Dive: Grok 4.7 is weirdly* good at electrical engineering

Most model launches lead with coding. The new Grok 4.7’s most unique result may be electrical engineering.

On xAI’s EEBench evaluation, Grok 4.7 scored 64.0%, up from Grok 4.6’s 53.0%. In xAI’s same table, Fable 5.1 scored 56.4% and GPT-5.6 Sol scored 39.4%.

That does not mean Grok suddenly wins every engineering problem. It does mean there may be a real specialization worth testing instead of picking one “best model” for everything.

Try it on a real circuit-debugging, schematic-review, or design task you can independently verify. Compare it with your usual frontier model, then judge accuracy, useful intermediate reasoning, and how much correction it needs.

The broader lesson: model selection is getting weirdly domain-specific. Your best coding model may not be your best electrical engineer.

*It’s actually not weird at all because SpaceX is VERY good at E.E!

New from The Neuron: AI Explained

Watch: OpenClaw 2.0 with Chief Architect Vincent Koc

Want to finally understand what OpenClaw actually is? We went live with OpenClaw Chief Architect Vincent Koc to talk about OpenClaw 2.0 and how the open-source personal AI assistant is evolving into a broader agent platform.

If you’ve been curious about running your own agents, connecting them to your computer and cloud tools, or just figuring out why everyone keeps talking about the lobster, watch the replay here.

A Cat’s Commentary

Btw: We just launched a robotics newsletter! Sign up for it here.