Meta dropped its Q1 2026 earnings this week and the numbers tell two very different stories depending on where you look. The core social media business is doing extremely well. Revenue hit $56.3 billion, up 33% year over year, and net income came in at $26.8 billion, a 61% jump compared to the same period last year. By any normal measure, that is a remarkable quarter.

Then there is Reality Labs, the division responsible for Meta's VR headsets, AR glasses, and all things metaverse. It lost another $4 billion this quarter, which sounds alarming until you look at the bigger picture; over 21 quarterly earnings reports since 2021, Reality Labs has lost a total of $83.5 billion, averaging roughly $4 billion in losses every single quarter. At this point it is less a red flag and more just a line item Meta's investors have learned to brace for.

Meta has projected it will spend between $125 billion and $145 billion in 2026, most of it going toward AI infrastructure. That is significantly higher than analysts expected and nearly double what the company projected spending just a year ago. Mark pointed to higher component costs, particularly memory pricing, as the main driver. When one investor asked about 2027 spending plans, Meta's CFO gave an unusually candid answer, saying the company keeps underestimating how much compute it actually needs.

Google is rolling out Gemini to cars with Google built-in, replacing the existing Google Assistant with a significantly more capable AI. The rollout starts in the US with English support, but what makes this particularly interesting is that it is not just for new cars. Existing vehicles that support Google built-in can receive the upgrade through a software update, meaning millions of drivers could already be eligible.

General Motors was the first automaker to confirm the news, announcing that Gemini is coming to around 4 million vehicles from model year 2022 and newer across Cadillac, Chevrolet, Buick, and GMC. Google's own announcement did not name specific automakers, which suggests the rollout extends well beyond GM and will cover a much wider slice of compatible vehicles.

Rather than issuing specific commands and hoping the car understands them, drivers can speak more naturally. You could ask Gemini to find a highly rated restaurant with outdoor seating along your route, and it pulls suggestions from Google Maps, answers follow-up questions about parking, and adjusts for dietary preferences, all in one continuous conversation. It can also handle climate controls, directions, music, vehicle information, message summaries, and hands-free replies without you needing to touch anything.

There is also a beta feature called Gemini Live, which opens up more open-ended conversations while you drive. Tap a button or say "Hey Google, let's talk" and you can brainstorm, learn something new, or just have a back-and-forth on whatever is on your mind during a long commute.

Drivers signed into their Google accounts in compatible vehicles will see an option to upgrade. Google says it plans to expand language and regional support over time, with deeper integration with Gmail, Google Calendar, and Google Home coming in future updates

Lovable, the vibe coding platform that recently crossed $400 million in annual revenue, has launched its mobile app on both iOS and Android. Your ideas do not wait for you to sit down at a desk. They show up on the bus, in the coffee line, or at 2am when you should definitely be sleeping. Now you can act on them directly.

The app is free to download on both the Play Store and the App Store, and lets users build web apps and websites using text or voice prompts. Once you describe what you want, Lovable's AI agent works through it autonomously in the background while you get on with other things. You can also queue up multiple prompts at once, switch between your phone and laptop without losing progress, and get a notification when a build is ready to review.

The mobile app is not a full replacement for the desktop version, it is more of an extension of it. Users can prompt ideas, preview results, and iterate while switching seamlessly between the mobile app and the web platform. On iOS specifically, generated app previews open in a browser rather than inside the app itself, which is how Lovable navigated Apple's recent crackdown on vibe coding tools.

Apple recently blocked updates to popular vibe coding tools including Replit and Vibecode for violations of its developer guidelines.The core issue is that apps generating and executing code after the App Store review process raises security concerns Apple is not comfortable with. Lovable sidestepped this by keeping outputs web-based, staying within the rules while preserving what makes the product useful.

Subscription plans on iOS range from $9.99 to $79.99 per month. For a platform already serving 8 million users and counting, putting the product in people's pockets is a natural next move.

Meta has acquired Assured Robot Intelligence, a humanoid robotics startup whose entire team, including its two co-founders, will join Meta's Superintelligence Labs research division. Financial terms were not disclosed. The startup was building foundation models that allow humanoid robots to understand and adapt to human behaviour in complex, real-world environments, essentially teaching robots to perform physical tasks like household chores by learning from their surroundings rather than following fixed instructions.

The two co-founders bring serious credentials. Xiaolong Wang comes from Nvidia's research team and was previously an associate professor at UC San Diego. Lerrel Pinto taught at NYU and co-founded Fauna Robotics, a startup building kid-sized humanoid robots that Amazon acquired just last month. The fact that a former Fauna co-founder is now joining Meta, shortly after Amazon picked up the other half of that story, says a lot about how fast the robotics talent market is moving right now.

Meta researchers have been working on humanoid robotics for years, and a leaked memo from a year ago discussed the company's ambitions to build a consumer-facing humanoid robot, including both AI models and hardware. Whether that ever becomes a product people can buy is still an open question, but the acquisition gives Meta a team specifically capable of pushing those ambitions forward.

Many AI researchers now believe that reaching artificial general intelligence will require training models in the physical world, where a robot learns through direct interaction rather than from data alone.

Build A Real Agent Team. Agentic Engine is the self-hosted agentic enterprise platform that detects, decides and acts.