The Internet Is Starting to Do Things on Its Own — And It’s Been Planned Since 2001
AI is no longer just answering questions. It’s booking appointments, filling out forms, and browsing websites without anyone touching a keyboard. Today’s stories trace a thread from a 25-year-old idea finally coming to life, to the developer tools quietly powering that shift, to a startup that just raised $62.5 million betting it can predict what humans do next.
An Old Dream for the Web Is Finally Working
The web was designed for humans to read. Back in 2001, Tim Berners-Lee — the person who invented the World Wide Web — had a different vision. He imagined software agents, programs that could read a webpage, understand what it meant, and take action without a person clicking anything. He called it the Semantic Web. It never quite worked, because the technology wasn’t ready.
According to MIT Technology Review, that’s changing now. Modern AI systems can browse a site, interpret what they find, and complete tasks on your behalf. Book a flight, compare prices, submit a form — all without you doing anything except asking.
Think of it like the difference between giving someone a map and giving them a car with a driver. The old web gave you information. The agentic web — where AI agents act on your behalf — takes that information and does something with it.
For everyday people, this means AI assistants that don’t just answer “what’s the cheapest flight to Chicago?” but actually go buy the ticket. That’s useful. It also raises real questions about what these agents can access, what mistakes they might make, and who’s responsible when things go wrong.
Why this matters: We’re moving from AI as a search tool to AI as an actor. That’s a significant shift in how people will interact with technology — and in how much they’ll need to trust it.
“Semantic Web concept from 2001 is now becoming reality as AI systems grow capable.”
The Plumbing Behind AI Agents Just Got an Upgrade
LangChain is a popular toolkit that developers use to build AI-powered applications — think of it as a set of pre-built pieces that let programmers snap together chatbots, AI agents, and automated workflows without writing everything from scratch. This week, LangChain shipped two updates: version 1.4.8 of its main library and version 1.7.4 of langchain-core, which is the foundational layer everything else runs on.
The 1.7.4 core release fixes a bug where data was accidentally being changed in ways developers didn’t intend — a small but important correction. The broader 1.4.8 release carries a larger piece of news buried in the changelog: starting with the next major versions, LangChain will no longer support Python 3.10. Python is the programming language most AI tools are built with, and version numbers matter because newer versions are faster and more secure.
“Python 3.10 will no longer be supported starting with langchain-core 1.8.0 and langchain 1.5.0.”
Dropping older Python support is a standard housekeeping move, but it signals that the LangChain team is pushing forward rather than maintaining backward compatibility indefinitely. Developers who haven’t upgraded their Python version will need to do so before the next major release.
For non-developers, this is mostly invisible infrastructure news. The applications you use — AI writing assistants, customer service bots, automated research tools — often run on frameworks like LangChain. Keeping that foundation current is what makes those tools more reliable over time.
Why this matters: The AI agent tools people are building today depend on a healthy underlying ecosystem. These updates are how that ecosystem stays stable.
Also Happening in AI
OpenAI quietly released a maintenance update to its Python library, fixing a WebSocket connection issue — WebSockets are the technology that lets applications maintain a live, two-way connection with an AI rather than making one-off requests. Separately, a startup called 42AI emerged from stealth this week with $62.5 million in funding, claiming its AI can predict human behavior using something it calls a “world model” — an AI system trained to simulate how people think and act, not just respond to text. OpenAI is also reportedly merging ChatGPT’s separate Chat and Work tabs into a single unified interface, a small UX change that reflects how the line between personal and professional AI use is blurring. And MIT Technology Review published a piece arguing that developers should stop optimizing AI for token efficiency — tokens being the small chunks of text AI processes — and start measuring it by what actually gets built and shipped.
What to Watch
The agentic web story is the one to follow closely. As AI systems gain the ability to act on the web independently, the companies that control access — browsers, operating systems, app stores — will face new pressure over what agents are allowed to do. Watch for early policy moves from Google and Apple in the next few months. The 42AI “human behavior prediction” claim also deserves scrutiny: what data it was trained on, and what its actual accuracy looks like in practice, will tell us whether this is a genuine tool or an ambitious pitch.