Welcome, humans.

Monday brings good tidings of a new trend: people using AI to transform pictures of their cats into everything from what they’d look like as people to Pink Floyd album covers. Purrrfect!

This makes us think of the Chinese robot that can play soccer. Why? Honestly we just wanted to share and needed a transition. We’ll try harder next time, we promise!  

But while we’re on the subject of robots, check out this robot boxing a human as it preps for the Robot Olympics:

Though TBH, of all the robot use-cases we’ve seen, the most impressive one might be this robo arm that can peel the film off an adhesive tape. That would’ve taken us like 100x longer, and with plenty of sticky tape residue left to scrape off.

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

• Google’s new vibe coding app is legit.

• Figure sought $1.5B raise at $39B+ valuation for its humanoid robots.

• Apple maintained its plans for a fall Siri release.

• AI reasoning model benchmarking costs surged.

AI agents in action, example #1: Google’s Firebase Studio is an AI agent for making apps.

Yesterday we ran a story on how to automate anything, trying to explain “agentic workflows” and agents (and the difference between those two).

Today, let’s check out an actual agent in the form of Google’s newly launched Firebase Studio. Think of it as an agent living in your browser, designed to build apps for you.

But instead of just performing one task at a time, this agent can handle a whole sequence of steps based on your high-level instructions

It works like other “vibe coding” platforms that build apps from prompts like Lovable or Bolt, but potentially much beefier. 

Here’s how it works: 

• You feed Firebase Studio a prompt, or even a sketch, with your app idea.

  • Example: “Make me a simple to-do list with login.”

• It taps into its AI brain (Gemini, naturally) to instantly spin up a working prototype.

• It wires up the code using Firebase, and can even add AI features with Genkit.

The integration with Genkit might be the killer feature: This means you can use AI to build apps with AI. Which is, like, very meta—but also very “agent” friendly. For example, you can…

• Easily swap in and out different AI models.

• Chain multiple AI together into complex workflows (example: analyze image > generate recipe > create image of recipe).

• Visually test, debug, and monitor the AI flows to see what they’re really up to.

Why this matters: Google getting into the vibe-coding trend makes perfect sense. Vibe coding lowers the barrier for anyone to start building software, which creates a new market for their Gemini AI models (and ecosystem of code-related resources).

If you’ve ever stared at a brilliant app idea sketched out on a sticky note on your desk, laying dormant, waiting for the day you finally start learning how to code, these tools are for you… Just think, you could be this guy in no time!

Our take: We’ve been saying this a lot lately, but Google is really cookin’ atm, as they basically have the best video model and the best language model available right now. Firebase Studio is yet another example of their dominance.

Which begs the question: Have you made the switch over to Gemini 2.5? Why or why not? And if not, what AI model are you using most right now?

Anyway, we had a blast building with Firebase Studio over the weekend—and we think you will, too. We broke down exactly how it works, compared it to Lovable / Bolt, and included a mini-tutorial you can follow, too. Check it out on the website here.

FROM OUR PARTNERS

When your AI needs the best ears in the business... 👂

Frustrated when voice AI constantly misunderstands you? Speechmatics fixed that.

While others rush to make AI talk, Speechmatics has solved what matters first: making it truly listen. 

Their real-time speech tech delivers 90%+ accuracy in under one second across 55+ languages, diverse accents, and dialects – a full 25% more accurate than competitors, even in noisy environments.

Whether it's AI assistants, customer service, or medical transcription, Speechmatics ensures AI catches every word the first time.

No more “can you repeat that?”—just AI that keeps up, not catches up.

Try Speechmatics at no cost and hear the difference.

Prompt Tip of the Day

When using general web search with Claude and you need to look up a list of things, ask your web searcher to search for each topic one a time, stopping after every four items and then prompt you to continue. Four is about the max these searches can do without getting confused or stopping halfway through.

Example:

Treats To Try.

1. Cognee gives AI agents structured memory through knowledge graphs in 5 lines of code, connecting relationships between facts for more accurate responses—free (code, explainer vid).

2. Prompt Catalyst is yet another prompt crafter, but this one seems tailored to helping with video and image generating specifically—check this adorbs demo.

3. Autoread automatically responds to your iMessages using custom prompts, letting you ghost responsibly while appearing productive when you need a break from your phone—free to try.

4. Hebbia can search massive documents and help answer questions, and is quite popular with lawyers because of it (raised $130M).

5. Photoroom helps you create professional product photos and marketing assets with AI-powered tools that remove backgrounds, retouch images, and generate creative—free to try, then $7.50/mo with yearly plan.

6. Hera turns text into customizable motion graphics instantly—free to try.

7. Andi is a privacy-focused AI search engine that gives you direct answers without ads or tracking—free to try.

See our top 51 AI Tools for Business here!

Around the Horn.

• Veo 2, Google’s state of the art video generator, may soon roll out widely inside AI Studio, as some users have reported getting early access.

• Apple reportedly still plans to release the new Siri this fall—meanwhile, a new report from the NYT revealed that Apple’s AI troubles go way back to 2023, when the CFO wouldn’t approve the AI team for more NVIDIA chips. 

• Allie K. Miller put together a great explainer on X that covers all the ins and outs of ChatGPT’s new memory feature.

• Artificial Analysis says AI reasoning models made benchmarking prohibitively expensive in 2025, with the company spending $5K+ to test twelve reasoning models compared to $2.4K for eighty non-reasoning models.

• Figure wants raise $1.5B at a $39B+ valuation despite having no revenue last year and only a few dozen robots in production.

• Here are the top AI papers from last week according to the NLP Newsletter—our fave is Browsecomp, which is OpenAI’s new benchmark for web agents.

FROM OUR PARTNERS

Monetizing AI Webinar: How to get it right

Make your pricing your competitive advantage. Join pricing experts from 49 Palms Ventures and Metronome CEO Scott Woody for a live discussion on how to design effective, sustainable pricing for your AI products. 

Register for the webinar today.

Monday Meme

A Cat's Commentary.