😸 June 17 (Wednesday)

Welcome, humans.

The tide might be turning on the anti-AI crowd dominating the creative community. The tools are getting better, and as a result, people are taking more serious shots on goal to create things with them:

That lines up with Adobe’s latest creator data: 87% of creators who use or have tried creative AI say it has helped grow their business or audience. The catch: creators still want control, authorship, and the final call.

Here’s what happened in AI today:

• 😺 Story 1

• 📰 Alibaba launched Qwen robot models for physical AI.

• 📰 China pushed humanoid robots from demos into work.

• 📰 SubQ showed a 12M-token long-context model report.

• 📖 Nate B Jones broke down the AI bubble debate.

…and a whole lot more that you can read about here(hyperlink bold text w/ link).

Hey: Want to reach 700,000+ AI-hungry readers? Advertise with us! 

P.S: Love robots? We’re starting a new robotics newsletter! Sign up early here.

🙀 SpaceX Is Buying Cursor For $60B

The hottest coding app in AI just became part of Elon’s trillion-dollar rocket empire.

That’s right: SpaceX will acquire Cursor, the AI coding startup behind the editor used by millions of developers, in a $60B all-stock deal. The deal is expected to close in Q3, according to Axios, and Business Insider reports it comes days after SpaceX’s record IPO.

Here's what happened:

• Cursor’s parent company, Anysphere, is being acquired by SpaceX for $60B.

• The deal is being paid in SpaceX shares, which makes SpaceX’s sky-high public valuation part of the acquisition currency.

• Cursor reportedly raised $3.38B since its 2022 founding from investors including Thrive, a16z, the OpenAI Startup Fund, and Nvidia.

• Business Insider reported that Cursor’s annualized revenue passed $1B after growing 10x in under a year.

And Cursor is also teasing a much bigger model push. At its Compile event, Morgan Linton shared video of Cursor announcing a new model (Composer 3), while Nick Dobos said it is in the same size class as Claude Opus and GPT-5.5, trained from scratch with no Kimi (open Chinese model) base, built with 10-20x more compute than Composer 1, and expected in the next couple of weeks. Ray Fernando added that the model is 1.5T+ parameters, retrained on 100k+ GPUs, and aimed at intelligence beyond coding.

That is the real bet. Cursor is no longer only a smarter code editor. It is becoming the workflow layer for software teams where humans and agents write, review, merge, and ship code together. And if those event posts are right, the Anysphere SpaceX combo could become a legitimate fourth leg to the frontier model lab four-legged race, and not just the app sitting on top of other people’s models. Many businesses are warming to this idea now after the whole Fable 5 debacle.

Why this matters: Coding agents are now strategic infrastructure. For SpaceX, owning Cursor could mean faster internal software development across rockets, satellites, autonomy, manufacturing, and Grok (and less relying on ever more unreliable Claude models… except for revenue for SpaceX datacenters).

For regular people (who we affectionately call normies), you might not care about yet another serious coding model, but remember this: eventually, all models will be coding models, and you will just ask them for things, and they will build them for you on the spot. So you want lots of competition to make this affordable for you to use without having to pay enterprise prices.  

FROM OUR PARTNERS

Governance workflows for AI agents

Production AI failures rarely come from a single bad model response. They slip through when evaluations, tests, traces, and approvals are scattered across disconnected tools and teams.

This practical guide shows how leading organizations are building governance workflows for AI agents without slowing development. Learn how to run automated evaluations, organize human reviews, investigate failures with complete traceability, and create evidence-backed release decisions.

Whether you're building customer-facing copilots or internal agents, you'll get a framework for making AI systems more reliable, auditable, and production-ready.

Get the guide to see the workflow in action.

🎓 AI Skill of the Day: Want better AI output? Give the model a leading word (or Leitwörter)

Matt Pocock shared a useful concept for guiding AI models which he apparently gleaned from literary theory: leitwörter, translated to repeated “leading words” that anchor meaning. In AI skills, a leitwort is a phrase the agent can reuse to guide its own behavior.

Example: Matt’s /teach skill uses a phrase called “zone of proximal development,” an education term describing the ideal state where a learner should feel challenged but not overwhelmed. (Side note: it’s a sick skill, and Elvis Saravia’s DAIR walkthrough shows how to use that /teach pattern to turn an AI assistant into a structured tutor; Matt also has lots of skills like these you can get here… check out both!).

Basically, he says a good leading phrase compresses a whole behavior into a reusable handle, and when he repeats it 2-3 times in a skill, he’s even seen the agent refer to the phrase in its own thinking phrases, guiding the behavior.

So pick one phrase that carries the whole behavior you want, then use that phrase as an operating principle in the prompt / skill.

Here’s a Matt Pocock-inspired prompt to apply this:

Use [Your LEITWORT here] as your operating principle for this task.

By [Your LEITWORT here], I mean: [simple definition of the behavior you want].

Apply that principle while you work. Before giving the final answer, check whether the output follows [LEITWORT] and revise once if needed.

Task: [paste your task here]

Context: [paste relevant context here]

Output format: [describe the format you want]

Good prompting isn’t always more instruction, or perfect formatting. It’s finding the right phrase that makes the model instruct itself.

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

🍪 Treats to Try

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

1. *Before you host another AI strategy meeting, build the prototype first. Dell Pro Max with GB10 gives teams a serious desktop AI launchpad. Check it out.

2. Skywork Design turns a simple prompt into consistent website pages, product flows, dashboards, and publishable designs you can edit on an infinite canvas, no pricing details.

3. Firecrawl now lets you search, scrape, interact with web pages, and parse PDFs into clean markdown with no API key required until you scale; free to try.

4. MDN MCP server gives coding agents accurate MDN docs and browser compatibility data inside tools like VS Code, Cursor, and Claude Code, free to try.

5. fal.ai FLUX.2 klein realtime helps you build low-latency image-to-image workflows for interactive creative apps.

6. GLM Coding Plan gives developers GLM-5.2 and GLM-5-Turbo inside agent and IDE workflows like Claude Code, Kilo Code, Cline, and OpenCode; plans start at $18/month.

7. Cursor Origin gives agent-heavy coding teams a git forge built for code that now moves faster than old review workflows.

8. Step 3.7 Flash is a high-efficiency multimodal model on OpenRouter with a 256K-token context window, priced at $0.20 per million input tokens and $1.15 per million output tokens.

9. Pica gives you a native Mac font manager with logo previews, collections, OpenType support, and watch folders.

📰 Around the Horn

1. Sensor Tower said ChatGPT hit 1B mobile MAUs while AI assistants reshaped shopping referrals, retailer apps, and early ChatGPT ads.

2. OpenAI released Deployment Simulation, which tests candidate models on de-identified real conversation patterns to predict risky behavior before launch.

3. Google launched Android 17 with AppFunctions, Bubble Bar multitasking, device handoff, post-quantum security, and more Gemini features.

4. Z.ai released GLM-5.2, an open-weights model with a 1M-token context window and strong long-horizon coding results.

5. CoreWeave set a new MLPerf record by training DeepSeek-V3 671B in about two minutes on 8,192 NVIDIA GB300 GPUs.

6. Anthropic paused token-based billing for its Claude Agent SDK after backlash from heavy users.

7. Alibaba launched Qwen Robot Suite, a set of models for robot navigation, object manipulation, and world prediction as the company pushes Qwen beyond chat and into “physical world intelligence.”

8. China set a 2026 plan to move more than 10,000 humanoid robots into real jobs across factories, logistics, retail, healthcare, inspection, and emergency response.

9. SubQ finally released the technical report for SubQ 1.1 Small, a long-context model report claiming near-perfect retrieval up to 12M tokens and 64.5x less compute than dense attention at 1M tokens.

FROM OUR PARTNERS

This is what real AI implementation looks like.

Venture needed to scale operations without adding headcount. Creai identified 75 opportunities, prioritized 11 initiatives, and deployed 7 systems in production. The result: $30M USD in annual returns and 90% automated resolution.

See how we did it. →

📖 Midweek Wisdom

• Epoch AI warned that AI capex across Microsoft, Amazon, Alphabet, Meta, and Oracle is on pace to exceed operating cash flow by Q3 2026.

• François Chollet argued that open AI (as in, AI designed and released to be open for everyone to benefit, not the company)’s path forward depends on radically better training efficiency, especially around data requirements.

• SemiAnalysis broke down why reinforcement learning systems bottleneck when trainer and generator throughput fall out of sync.

• Microsoft Design explained its new AI-forward design system around Presence, Memory, Attention, and Shared Awareness.

• Ed Zitron reported that OpenAI’s 2025 losses rose nearly 8x to $38.53B on $13.07B in revenue, with costs driven by R&D, Microsoft payments, and a one-time accounting charge tied to its nonprofit-to-for-profit conversion.

• Hiten Shah argued that GitHub is becoming the place where non-engineers’ judgment enters the software workflow earlier through plans, issues, and durable context agents can act on.

• Yining Hong explored whether AI can truly “discover,” arguing the real issue is who sets the evaluation criteria when systems generate, test, and retain new ideas.

• Windows 11 has a pile of underused productivity and AI tools, including Copilot Vision, Recall, Live Captions, Snipping Tool, Paint, passkeys, and Snap layouts.

• A new corporate phrase has entered the chat: “tokenminimizing”; according to The Information, AT&T has started throttling some employees’ AI usage as companies realize those magical productivity boosts come with very real model bills.

• Your $20 AI Plan Costs Them Thousands. That’s Not The Bubble: Nate B Jones breaks the AI bubble debate into actual business segments, which is a lot more useful than yelling “pets.com” every time someone mentions data centers.

A Cat’s Commentary

P.S: Before you go… have you subscribed to our YouTube Channel? If not, can you?