OpenAI v Anthropic Gets Spicy; Lovable Under Assault? AI Trust Low

Today's AI Outlook: 🌤️

When The Boss Is A Bot

An AI agent named Luna is no longer trapped in demo-land. Andon Labs handed it a real retail store in San Francisco, a credit card and one job: make money. Luna came up with the boutique concept, posted jobs, interviewed candidates, and managed store operations, which makes this less “chatbot with opinions” and more “entry-level CEO with dropdown-menu issues.” The early results are exactly what this phase of AI looks like: surprisingly capable, occasionally absurd and very real.

Why it matters

This is the clearest sign yet that AI may move into management and coordination before it fully replaces frontline labor. The software still makes goofy mistakes, but the direction of travel is getting harder to ignore. When an agent can lease space, hire people, and run schedules, the line between tool and operator starts to blur fast.

The Deets: Luna reportedly ran on Claude Sonnet 4.6 for reasoning and Gemini 3.1 Flash-Lite Preview for voice, using security camera screenshots to monitor the store. It handled hiring over Zoom with the camera off, then fumbled at least two very human-sounding mistakes: choosing Afghanistan by accident in a TaskRabbit dropdown and mangling the opening-weekend staff schedule. In other words, it is already managing like a real boss in at least one respect.

Key takeaway

AI agents are graduating from toy tasks to messy real-world responsibility, and the comedy of errors is no longer the main story. The main story is that the experiment happened at all.

🧩 Jargon Buster - Agentic AI: An AI system that does more than answer questions. It can take actions, make choices, and move through a workflow with limited human supervision.

⚔️ Power Plays

Platform Wars, With Extra Corporate Spice

The OpenAI-Anthropic rivalry is starting to sound more like a knife fight in a boardroom.

A leaked OpenAI memo reportedly dismisses Anthropic’s $30B run rate as inflated, calls it a single-product company, and frames the whole battle as a platform war. Meanwhile, AI Secret paints the bigger arc: both companies are scrambling to own the full stack, from models to agents to the software layer built on top of them.

Why it matters

This fight is no longer about who has the smartest model on benchmark day. It is about who controls the customer relationship, the workflows, and eventually the businesses built on top. That shift matters because platform wars tend to crush the middle. If the model providers absorb more of the app layer, a lot of AI wrappers start looking like temporary housing.

The Deets

OpenAI’s memo reportedly leans on its Amazon partnership as a path to more enterprise growth and less dependence on Microsoft, while taking shots at Anthropic’s compute constraints and availability issues. At the same time, Anthropic appears to be expanding Claude into a broader product universe that reaches across legal, finance, security, automation, and app building. The common theme is simple: nobody wants to be “just the model” anymore.

Key takeaway

The next AI winners may not be the labs with the best chatbot. They may be the ones that turn models into the default operating layer for work.

🧩 Jargon Buster - Platform War: A competitive fight to control the ecosystem around a product, including apps, users, partners, and workflows, not just the core technology itself.

Linux Lets AI In, With Caveats

One of the most stubbornly skeptical corners of tech has made its peace with AI code, at least on paper. Linux kernel maintainers now allow AI-assisted contributions, but the rules are blunt: AI tools cannot sign patches, and the human contributor keeps full legal, quality, and security responsibility.

Why it matters

This is a symbolic surrender from one of the world’s most consequential software communities. AI is now inside infrastructure that runs billions of machines, but nobody is pretending that trust comes bundled with autocomplete.

The Deets

Developers must disclose AI assistance with an Assisted-by tag, which creates a paper trail without giving the model any accountability theater. That tells you a lot about where the industry is landing: yes to speed, no to outsourcing blame. The code may come faster, but the human still owns the consequences.

Key takeaway

AI coding has crossed from novelty to accepted workflow, but responsibility still stubbornly remains a human job.

🧩 Jargon Buster - Assisted-by Tag: A disclosure label showing that AI helped produce code, while the human submitter remains responsible for what gets shipped.

🛠️ Tools & Products

Claude Going After Lovable, Vibers?

Anthropic looks like it is testing a vibe-coding feature inside Claude that would let users build simple apps from prompts, based on leaked screenshots shared on X.

The images suggest users could generate things like AI chatbots, photo albums and landing pages directly inside the chat experience. If that rolls out, Claude would start looking less like a chatbot and more like a full-stack app builder.

That would put Anthropic in more direct competition with startups like Lovable, the Stockholm-based company that has become one of Europe’s breakout names in no-code AI app creation. Lovable has been riding the surge in demand for tools that let non-engineers spin up software with plain English. Translation: the “just tell the machine what you want” market is getting crowded fast.

Why it matters

This is another sign that the big AI labs are moving up the stack. They are not just selling models anymore, rather they are chasing the products built on top of them. For startups like Lovable, the challenge is obvious: it is one thing to outrun other startups, and another to fend off a company with Anthropic’s capital, distribution and model access. The cozy “AI tools ecosystem” is starting to look more like a knife fight in a hoodie.

The Deets

Lovable was founded in 2023 by Anton Osika and Fabian Hedin and has quickly become one of Europe’s hottest AI startups. In December, it raised $330M at a $6.6B valuation, more than tripling its valuation from July 2025. Its backers include Accel, Creandum, Evantic, CapitalG and Anthology fund.

Lovable’s team has already been bracing for pressure from larger players. Last month, head of growth Elena Verna said on the 20VC podcast that Big Tech, not fellow vibe-coding startups, is the real threat. Meanwhile, Osika recently said on X that Lovable is looking for more teams and startups to join the company, and hired Théo Daniellot, who previously worked at Revolut and Ledger, to lead M&A.

Anthropic’s move also fits a pattern. A few months ago, it launched a legal tool that rattled Europe’s legal tech startups. Analysts warned at the time that today’s AI encroachment might be legal, but tomorrow it could be sales, marketing or finance. Now it may be app building’s turn to sweat.

Key takeaway

The AI giants are are climbing into the driver’s seat. If Claude becomes an app builder, startups like Lovable will need to prove they can out-execute the labs that power the whole category.

🧩 Jargon Buster - Vibe coding: Building software by describing what you want in plain language, while AI handles much of the coding and app setup behind the scenes.

Google Puts AI in Your Pocket... Sans Internet Connection

Google’s latest mobile AI setup is aimed at a very specific pleasure: running a solid model on your phone without accounts, subscriptions or an internet connection after setup. The pitch is simple and powerful. Download the app, grab the model, and start chatting locally.

Why it matters

Local AI is becoming less of a hacker flex and more of a mainstream product experience. That means more privacy, less latency, and fewer excuses for why useful AI has to live inside a cloud bill.

The Deets

Google AI Edge Gallery is where users can download a model such as Gemma 4 E2B, enable Thinking, and run chats directly on-device. The app does not save chat history, which is either a flaw or a privacy feature depending on how your week is going. It also includes extra Agent Skills like restaurant roulette, Wikipedia lookup, maps, and QR generation.

Key takeaway

AI on-device is quietly becoming normal, and that matters because normal is where adoption gets dangerous for incumbents and delightful for users.

🧩 Jargon Buster - On-Device AI: AI that runs directly on your phone or computer instead of sending requests to remote servers.

💰 Funding & Startups

Figure Buys Compute Like A Drunken Robot

Figure AI and CEO Brett Adcock’s startup Hark have reportedly secured an entire datacenter of NVIDIA B200 Blackwell GPUs to scale robot learning and multimodal model training.

Why it matters

Physical AI is becoming a compute sport. Training robots to understand motion, physics and messy environments is expensive, and the companies with real infrastructure are putting distance between themselves and everyone still pitching robotics with a nice demo reel.

The Deets

Robotics Herald says the compute will support both robot learning and physics prediction ahead of broader production efforts. It also underscores a bigger point: robotics startups are starting to look a lot like frontier AI labs, with giant hardware appetites and serious capital intensity.

Key takeaway

In robotics, ambition is increasingly measured in datacenter acreage.

🧩 Jargon Buster - Multimodal Model: An AI model that can work across different types of data, such as text, images, video, audio, or sensor streams.

🧪 Research & Models

AI Adoption Is Up, Trust Is Not

Stanford HAI’s 2026 AI Index lands with the kind of split-screen story the industry keeps trying to outrun. Global AI adoption has reached 53%, but public trust is stuck in the basement. Experts remain broadly upbeat about the job impact of AI, while the public looks at the same trend and sees a layoff with a better logo.

Why it matters

The AI story is now two stories. One is about acceleration, deployment, and economic upside. The other is about distrust, labor anxiety, and a public that does not believe the adults are steering.

The Deets

The report says nearly three-quarters of experts are optimistic about AI’s impact on jobs, versus just 23% of the public. The U.S. may lead in building AI, but it ranks only 24th in adoption at 28.3%, according to The Rundown’s summary. It also flags weakening trust in government management of AI-related change, plus pressure on younger developers as entry-level employment softens.

Key takeaway

AI does not have a capability problem. It has a legitimacy problem.

🧩 Jargon Buster - Adoption Rate: The share of people, companies, or markets actively using a technology, not just talking about it.

Robots Learn to Handle Cash, And “World Model” Debate Gets Messy

Generalist AI introduced Gen-1, a robotics model designed to handle real-world physical tasks like folding laundry, sorting objects, and handling cash. The company is also taking a swing at one of robotics’ favorite buzzphrases, arguing that “world models” and vision-language shortcuts are less useful than physics-native systems trained on actual interaction data.

Why it matters

Robotics is moving from impressive motion to useful manipulation, and the argument over how to get there is getting louder. This is not just a product launch. It is a vote in a deeper technical fight over what kind of model should power embodied AI.

The Deets

Robotics Herald reports Gen-1 was trained on human motion data rather than robot demonstrations and draws on more than 500,000 hours of real-world interaction data. The pitch is that physical common sense matters more than clever repackaging of existing language-and-vision systems. Whether that claim holds up long term, the direction is clear: robotics teams want models built for atoms, not just pixels.

Key takeaway

Physical AI is becoming its own discipline, with its own dogmas, data moats, and increasingly sharp elbows.

🧩 Jargon Buster - World Model: A system meant to predict how the environment behaves so a robot or AI can plan actions before taking them.

⚡ Quick Hits

Unitree Hit 10 m/s: Unitree’s H1 humanoid reportedly reached 10 m/s in a track test on April 11, matching the fastest recorded humanoid benchmark and turning the robot sprint race into a very real thing.

Robot Birds Are Doing Conservation Theater: Robotic sage grouse decoys are mimicking mating rituals in Grand Teton National Park to lure endangered birds back to safer breeding grounds. Strange plan, smart use case.

DroneDogs Are Guarding Corn: Bayer is deploying robotic dogs with AI and thermal cameras to patrol seed fields in Hawaii and California, filling security gaps and feeding incident data back to the cloud.

Underwater Robots Get a Coordinator: Thales introduced an AI-powered orchestration system for autonomous underwater robots in mine countermeasure missions, combining sonar analysis with multi-robot coordination.

Washington Has Formed a Robotics Commission: The Special Competitive Studies Project launched a National Security Commission on Robotics for Advanced Manufacturing, with reports due in late 2026 and March 2027.

OpenAI Made a Finance-Flavored Acquihire: OpenAI reportedly acquired AI finance startup Hiro in a move to deepen its capabilities in financial applications.

Vercel Is Riding the AI App Boom: AI Secret says Vercel is seeing rapid revenue growth from the surge in AI-generated apps and is signaling IPO readiness.

🧰 Tools of the Day

Scroll turns a knowledge base into an AI experience for employees and customers, which is a polite way of saying your docs might finally become useful.

Harvey Agents brings agent workflows to legal work like memos and diligence reports, continuing the strong industry trend of turning expensive billable hours into software.

Lovable Payments promises one-chat payment setup for AI-built apps, which sounds dangerously close to making shipping a product feel easy.

HeyGen CLI brings video generation to the terminal for builders who prefer prompts, pipelines, and not clicking around in dashboards.

Today’s Sources: The Rundown AI, AI Secret, Robotics Herald