Welcome, humans.

Want to see the best depiction of what its like to work with AI in 2025 vs 2023? Here you go:

If any of you have seen the movie Whiplash, you probably know where this is going… (worth rewatching this scene after you watch the above video to compare; he really nailed it).

Speaking of AI getting more capable, Chinese startup Moonshot AI just released Kimi K2—a 1-trillion-parameter open-source model that's absolutely crushing coding benchmarks (and had a literal moonshot moment as usage exploded on Openrouter).

This thing scores 65.8% on SWE-bench Verified (that's the test where AI actually fixes real GitHub bugs), beating GPT-4.1 by 11 points and DeepSeek V3 by a whopping 27 points.

Now get this: Kimi K2 is actually free to download and run yourself—no corporate overlords, no usage limits, just pure agentic AI that can write code, execute shell commands, and analyze data like your terminal grew a brain.

Try it yourself here on Openrouter.

Here’s what you need to know about AI today:

• Ex-Uber CEO Travis Kalanick is making AI burritos now.

• Tesla to launch Grok 4 in cars this week.

• The EU released new guidance on how to work with its AI Act framework.

• Meta acquired PlayAI while Google acqui-hired Windsurf’s team.

Travis Kalanick's robot army is coming for your lunch order.

Uber co-founder Travis Kalanick is back, and his new venture, Cloud Kitchens, is quietly rolling out the next phase of food automation. Forget ghost kitchens—we’re talking about fully robotic restaurants. Check out our full recap of the convo here.

On the All In Podcast this week, he revealed that his team has built a 60-square-foot machine that can assemble 300 custom bowls an hour for brands that look a lot like Chipotle or Sweetgreen.

The goal isn't just to make food faster; it's to fundamentally rewire the economics of the restaurant industry.

Here’s how the robot works:

• You order online, and the machine gets to work.

• A small carrier shuttles a bowl between dispensers that drop in precise amounts of every ingredient.

• It sauces the bowl, seals it with a lid, drops it in a bag with utensils, and sends it down a conveyor belt.

• Another robot picks up the sealed bag and places it in a smart locker, where your DoorDash driver scans their phone to pop it open.

This end-to-end system is already showing insane results. Kalanick says his customers are cutting their labor costs from a whopping 30% of revenue down to just 7-10% (8:43). Staff can prep ingredients, load the machine, and leave it to run autonomously for hours.

This is bigger than just automated assembly. 

• The vision, which Kalanick calls “autonomous burritos” (5:02), is to pair these robotic kitchens with self-driving delivery cars.

• The next step is tackling “state change”—the actual cooking of food.

• Imagine a line that grills a burger patty and drops it on a bun, feeding directly into the assembly robot (13:20).

• The more dispensers you add, the more dishes you can make, creating a future "internet food court" (16:53) where almost any meal is possible.

Why this matters: Travis argues this is a blueprint for the future of service industries. As personalized, robot-made meals become cheaper than groceries (15:06), entire industries from grocery retail to real estate will be forced to adapt.

Our take: The end game here, as one of the hosts mentions, is everyone getting their own personal robot chef. The giant 60 foot machine that Travis built is a far cry from Rosey the Robot in the Jetsons, but you apply the progress being made by Figure AI in humanoid robotics, and you start to see the possibility: yeah, eventually, we could all have a robot who learns on the job how to cook our favorite dish just like we like it.

FROM OUR PARTNERS

The AI Memory Economy

The future of AI isn't about who builds the smartest model; It's about who controls the memory.

Right now, your conversations with ChatGPT vanish into the corporate ether, your carefully curated prompts feed someone else's bottom line, and your digital brain gets lobotomized every time you start a new chat.

But imagine if your AI's memory was actually yours — portable, persistent, and profitable. What if the vector databases that store your AI's knowledge weren't locked away in some giant server farm, but sitting in your pocket like a digital sommelier.

This is the AI Memory Economy, and it's about to turn the entire industry on its head.

Cure Your AI’s Amnesia.

Prompt Tip of the Day

Yesterday, a confused Redditor posted one of those overwhelming “ChatGPT Prompting Cheat Sheets” filled with acronyms like R-A-I-N and F-L-O-W and asked for help. The community's advice? Ditch the acronyms and frameworks.

The top-voted response: “Using ChatGPT comes down to literally just talking. That's it.”

Instead of complex prompts, try simplifying things: “I need help writing a cover letter. I'm applying for marketing roles and have strong social media experience.”

And then give it all the context about what you’ve done and what the roles you’re applying for require, and let it cook. This is where “context engineering” comes in.

When you get a response, give it feedback on what you like and don’t like or what still needs work. And truly, for most things, that’s all that’s needed.  

We finally made it easy to talk to computers with natural language… so why are we overcomplicating things by talking like robots again?

Check out all of our Prompt Tips of the Day from June here. 

P.S: Completely new to AI? Start here!

Treats To Try.

*Asterisk = from our partners. Advertise in The Neuron here.

1. *Guidde turns your screen recordings into professional video tutorials with AI-generated step-by-step narration and voiceover in 100+ languages.

2. Endorphin lets you write end-to-end tests in plain English that automatically generate, validate, and fix themselves through AI—free to try.

3. BrowserOS runs AI agents locally in your browser to automate tasks while keeping your data private, offering a Chrome alternative with built-in automation—free.

4. Cactus lets you deploy and run AI like language models, vision language model, and text to speech models locally on smartphones with support for any GGUF model from Huggingface—open source, so free (read more) plus there’s an app!

5. Pixelesq autonomously designs, writes, and optimizes your website from documents or ideas while you maintain creative control—no pricing details.

6. Hello.cv creates your professional AI resume on a custom .cv domain to enhance your online visibility—free.

7. Need a name for your new dog? Someone created this Dog name generator in less than 24 hours entirely using AI (where’s the cat version, am I right?).

See our top 51 AI Tools for Business here!

Around the Horn.

• OpenAI indefinitely delayed its open-source AI model release due to safety concerns, which would have had similar capabilities to OpenAI's o-series, because releasing model weights would be irreversible, so they needed more time for red-teaming.

• Google hired AI coding startup Windsurf's top talent and acquired nonexclusive licensing rights to its tech for $2.4B after OpenAI's $3B acquisition attempt collapsed (reportedly because of the Microsoft of it all).

• Goldman Sachs became the first major bank to test Cognition's Devin AI agent, planning to deploy hundreds of instances to augment their existing 12K human developers.

• Meta acquired PlayAI, a California-based startup specializing in AI-generated voice cloning and speech synthesis technology.

• Amazon may invest more in Claude maker Anthropic in order to surpass Google’s investment in the company (for more control, we presume), stay competitive against OpenAI and Microsoft, and maybe even hold off an acquisition from Apple (if they’re even considering that… which they SHOULD).

• Tesla will bring Grok to its vehicles this week “at the latest” to serve as an advanced voice assistant that can handle everything from weather queries to dynamic route planning.

• The team behind Manus, the popular general purpose agent, has started recruiting in Singapore instead of China in order to gain access to global talent.

• SAG-AFTRA, the actors union, signed a new deal with video game studios so voice and motion capture actors gained consent for use of their digital likeness.

• The EU published an AI Code of Practice to help companies comply with AI Act obligations on safety, transparency and copyright. These include:

  • Companies must use a Model Documentation Form.

  • Provide information to downstream users within 14 days.

  • Implement copyright policies to only crawl lawfully accessible content.

  • And for advanced models like OpenAI's o3 and Anthropic's Claude 4 Opus, conduct systemic risk assessments and incident reporting.

• NVIDIA briefly became the first company to hit a $4T (trillion!!) market cap, which is the total value of all shares added up, due to demand for its AI chips.

• Rohan Paul shared the top 14 AI papers from last week in this easy to follow recap.

For the latest AI deep dives, check out our Explainer articles here!

ICYMI: Learn how to best wield AI in your business!

This week: An AI veteran who's interviewed 100+ leaders reveals the exact framework that took him from 4-hour workflows to 90-second automation. Plus, he drops truth bombs about how to avoid making AI projects that are just “bad decisions faster.” Listen now: YouTube | Spotify | Apple Podcasts

Monday Meme

A Cat's Commentary.