An AI Model Built Mostly by AI — and a Robot That Shops for You
Three stories landed today that, taken together, tell you something important about where AI is heading: the technology is starting to build itself, companies are betting enormous capital on staying in the race, and your next shopping trip might not require you at all.
A Startup Built a Coding AI Using Mostly AI
A company called Recursive has released a new coding assistant called Flint, and the story behind it is arguably more interesting than the product itself. According to TechCrunch, Flint was reportedly built primarily by AI systems rather than human programmers — which means AI wrote much of the code that powers an AI. Flint is also designed to run on a standard laptop, without needing a connection to a powerful remote server.
Think of it like a factory that starts building its own machines. Humans set up the original equipment and define the goals, but the assembly line increasingly runs itself. That’s the direction Recursive says it took with Flint, letting AI systems do the heavy lifting of actual construction.
For everyday people, the laptop angle matters most right away. Most capable AI tools today require you to send your data to someone else’s servers in the cloud. A model that runs locally means your code and queries stay on your own device. That’s a real privacy difference, especially for anyone working on sensitive projects.
Why this matters: If AI systems can reliably build other AI systems, development cycles could shrink dramatically. That changes how fast this whole field moves.
“Model was created mostly by AI systems rather than human programmers.”
Meta Is Creating New Shares to Fund Its AI Ambitions
Meta — the company behind Facebook, Instagram, and WhatsApp — is reportedly planning to issue a new class of shares specifically to fund its push into artificial intelligence, according to The Verge. These would be non-voting shares, meaning people who buy them get a financial stake in Meta’s future but no say in how the company is run. The plan is expected to kick off in 2027.
A non-voting share works like owning a piece of a business without getting a seat at the table. You can profit if the company grows, but the founders and existing voting shareholders still make all the calls. Meta’s CEO Mark Zuckerberg already controls the company through a dual-class share structure, so this move is really about raising money without diluting that control.
For regular people, this signals how expensive the AI race has become. Companies like Meta aren’t just spending from existing profits — they’re restructuring their entire financial architecture to pull in more capital. The implication is that staying competitive in AI requires so much investment that ordinary business operations can’t fund it alone.
Why this matters: When a company redesigns how it sells ownership to fund one specific bet, that tells you how seriously it takes the stakes. Meta is treating AI less like a product line and more like an existential priority.
“New non-voting shares for AI funding starting in 2027.”
Walmart Wants an AI Agent to Do Your Grocery Run
Walmart is reportedly launching an AI shopping agent — a piece of software that can browse its website and buy products on your behalf — according to The Verge. The system reportedly includes controls so you can review purchases before they go through and set a ceiling on how much it can spend. Walmart says its approach differs from how competitors like Amazon are handling AI shopping tools.
An AI agent is software that doesn’t just answer questions but takes actions in the world on your behalf. Instead of asking “what’s a good blender?” and then searching yourself, you’d tell the agent what you need and it would find options, compare them, and check out. The human stays in the loop at key decision points, but the browsing and selecting happen automatically.
If this rolls out broadly, the biggest change is time. Routine restocking — paper towels, pantry staples, household basics — could happen with minimal effort. The spending limit feature is a sensible guardrail, giving people a way to delegate without handing over unlimited access to their wallets.
Why this matters: AI moving from answering questions to completing purchases is a significant step. It shifts AI from a tool you consult to one that acts in your name.
“AI agent that can browse and buy products on Walmart’s website on your behalf.”
Also Happening in AI
On the developer side, several tool updates landed this week. OpenAI shipped two updates to its Python library — version 3.36.0 adds support for Google’s Vertex AI platform, which lets developers access OpenAI models through Google’s infrastructure, and 3.36.1 patches a bug that was blocking custom WebSocket connections. Anthropic, the company behind the Claude AI models, released version 2.0.0 of its Python SDK with breaking changes — meaning developers using it will need to update their code — followed quickly by 2.1.0, which adds new features for managing user accounts. LangChain, a widely used toolkit for building AI-powered applications, released version 1.4.7 with security patches and new capabilities for handling data. None of these changes affect most end users directly, but they reflect how rapidly the underlying plumbing of AI products is still being rebuilt.
What to Watch
Keep an eye on how quickly Walmart’s AI shopping agent expands beyond a pilot — and whether other major retailers announce similar tools in response. The Recursive story deserves follow-up too: independent researchers haven’t yet verified how much of Flint was genuinely AI-built, and that claim, if substantiated, would mark a meaningful threshold in AI self-development. When AI building AI moves from a headline to a routine practice, the pace of everything else in this field accelerates with it.