15 november 2026 · 14:25
ChatGPT Wallet: AI agents shop for you, LangChain hits $3B
OpenAI has launched ChatGPT Wallet, a feature that lets its AI agents select products, compare prices and complete purchases online for US Pro subscribers. In this episode of The State of Tech, we explain how the payment setup works, why the per-purchase fingerprint or face check matters, and what it could mean for European shoppers and regulators. We also cover LangChain's $300 million funding round, which values the AI agent startup at $3 billion.
Beluister deze aflevering:
Transcript
Samantha: Welcome to The State of Tech, The European Edition, Sunday November fifteen, 2026. I'm Samantha Lawrence.
Bob: And I'm Bob Russell. Today: ChatGPT gets a wallet so AI agents can shop for you, LangChain raises three hundred million dollars, Anthropic takes Claude into factory software, the Webb telescope finds ice older than its own star, and two things to check out: ElevenAI, the new workspace app from ElevenLabs, and the footage of SpaceX's Mechazilla tower catching a Starship. Let's start with your wallet, or rather, ChatGPT's.
OpenAI launches ChatGPT Wallet, letting AI agents buy things online on your behalf.
Samantha: OpenAI wants its AI to do your shopping, all the way to the checkout. The company has launched ChatGPT Wallet, which lets its AI agents, which OpenAI calls dots, actually buy things online for you. The agent picks the item, compares prices across supported shops and completes the purchase. For now it's rolling out to ChatGPT Pro subscribers in the United States. Until now, chatbots could recommend a product, but a human still had to type in the card number and click buy.
Bob: The payment side is where OpenAI has been careful. ChatGPT Wallet is built with Visa, Mastercard and American Express, and it uses single-use payment tokens. In plain terms: every purchase gets a one-off stand-in for your card number, so neither the shop nor the AI agent ever sees your real card details. And every single transaction needs your explicit approval with Face ID or a fingerprint before any money moves. The agent can fill the basket, but it can't spend a cent without your face or your finger.
Samantha: And that approval step carries a lot of weight. The pitch is convenience: tell the AI you need running shoes under a certain budget, and it does the searching and comparing. But an agent hunting across many shops could also pick the wrong size or the wrong product. Requiring a fingerprint or face scan for each purchase keeps the human as the final decision maker. The bottom line is that OpenAI is betting people will hand an AI the search and the comparison, as long as they keep the last click.
Bob: There's a market angle too. Visa, Mastercard and American Express all signing on shows the card networks want to be inside this new layer of shopping rather than bypassed by it. For online shops, it raises a new question: if an AI agent does the comparing, how clearly a machine can read your prices and product data may start to matter as much as your shop window. What the announcement doesn't say is which shopping platforms are supported, or how many, so the reach of this first version is still unclear.
Samantha: For Europe, this isn't available yet, and it may take a while. Regulators here are expected to look closely at agentic payment systems before any EU expansion, under Europe's data protection rules and the payment rules that already demand strong customer authentication, meaning an extra proof of identity for many online payments. OpenAI's biometric check per purchase is one form of such a step, but whether it satisfies European supervisors hasn't been tested. Key takeaway: European shoppers will likely watch American users try AI checkout first.
LangChain raises three hundred million dollars, valuing the AI agent startup at three billion.
Bob: Investors are betting big on the plumbing behind AI agents. LangChain has raised three hundred million dollars in a funding round led by CapitalG, putting the company's value at three billion dollars. Several other investment firms, including Sapphire Ventures, joined in. LangChain started in 2022 as an open-source toolkit, free software developers used to connect AI models to their own data and apps. It has since grown into a business selling tools to companies that want to build their own AI agents.
Samantha: That business is called LangSmith, and more than seven thousand organisations use it, including Nvidia, Toyota and Salesforce. LangSmith lets companies build AI agents, but also monitor them, test them, and set rules for what they're allowed to do. Think of it as a control room: you can see what your agent did and catch problems before they spread. As companies hand real tasks to software that acts on its own, LangChain is positioning that oversight as the product they need.
Bob: The other selling point is neutrality. LangSmith works regardless of which AI model or which cloud provider a company uses. So a business can switch between models from different suppliers without rebuilding everything, and keep control of its own data. That puts LangChain in an interesting spot: the big model makers would like customers to stay inside their own ecosystems, while LangChain is betting many companies prefer not to depend on one vendor. A three-billion-dollar valuation shows investors see room for that bet to grow.
Samantha: For European companies, that neutrality matters for a specific reason. The EU's AI Act asks organisations to keep human oversight over certain AI systems, and data sovereignty, keeping control over where your data lives and who can reach it, is a recurring concern here. Tools that log what agents do, and let a company stay independent from any single model provider, could help firms document that oversight. Whether that's enough for compliance still depends on how a company actually uses the tool.
Samantha: Quick interruption. If you listen to The State of Tech regularly, hit that like button and subscribe, that way you'll never miss an episode. Okay, moving on.
Anthropic launches Claude for Manufacturing, connecting its AI assistant to the software factories already run on.
Bob: Anthropic is taking its AI assistant Claude into the software that runs factories. The company has introduced Claude for Manufacturing, an add-on that connects Claude directly to industrial business software, including Autodesk Fusion, Oracle NetSuite and Infor. It comes with seventeen ready-made skills covering sales, quality management, supply chain and maintenance. The target is the information-heavy work around production, not the machines themselves, at least not yet.
Samantha: The examples are concrete. Claude can draft a job quote for a customer, evaluate bids from suppliers, and compile production reports from records the company already has. When equipment breaks down, it can analyse sensor readings to help find the cause. But it works strictly with a human in the loop: an employee has to explicitly approve before Claude carries out any action. So the AI prepares and drafts, and a person signs off. In a sector where a wrong order or a misread fault can stop a production line, that boundary is central to the offer.
Bob: There's a longer game here too. Anthropic is separately developing hardware standards that would eventually let its AI interface directly with physical machinery. That's not part of this launch, and no timeline has been given. But according to those plans, Claude would go from reading reports about the machines to one day talking to the machines themselves. For now, this is about the paperwork side of manufacturing: quotes, purchasing, quality records and maintenance logs.
Samantha: For Europe's industrial base, from German machine builders to Italian component makers, the approval step is relevant. EU safety standards expect human oversight, and a tool that can't act without a sign-off fits that expectation more easily than a fully autonomous system would. The opportunity flagged for European manufacturers is mainly in supply chain operations, things like comparing supplier bids. One open question: the announcement names only three software systems, so how well it fits the mix of tools in European factories remains to be seen.
The Webb telescope finds ice around a young star that is older than the star itself.
Bob: Some of the ice around a young star turns out to be older than the star. Astronomers using NASA's James Webb Space Telescope studied Beta Pictoris, a young star surrounded by a disk of dust and debris. In that disk they found water ice and frozen carbon dioxide with an amorphous structure, meaning the molecules are jumbled rather than arranged in a neat crystal. By analysing the light, they concluded this ice formed in the space between stars, before Beta Pictoris and its planets were born.
Samantha: That's striking because the birth of a star is violent, with intense radiation and heating. Fragile ice could easily be destroyed or transformed along the way, yet this ice survived in pristine form. And that matters for planets. Water and carbon dioxide are among the compounds that help form planetary atmospheres, and they play a role in prebiotic chemistry, the chemistry that could come before life. So some raw ingredients of future planets may be inherited from interstellar space, not only made locally.
Bob: The method is worth a moment. Spectral analysis means splitting light into its colours and looking for the fingerprints of specific molecules. Here, those fingerprints revealed not just which ices are present, but what state they're in, and that jumbled state is what gives away their age. What one observation can't answer is how common such ancient ice is around other young stars. That's the kind of follow-up question this result opens up for astronomers.
Samantha: For European astronomers working with Webb's spectral data, this is fundamental material. It helps explain how water and other building blocks that predate a star end up in the planets forming around it, a question that reaches across planetary systems throughout the galaxy. Key takeaway: some of the raw material of planets is older than their sun, and it can make it through the star's birth intact.
ElevenLabs unveils ElevenAI, an experimental app combining meeting transcription, dictation and automated follow-ups.
Bob: To close out, two things worth checking out today. First, voice tech company ElevenLabs has unveiled ElevenAI, an experimental workspace app for teams. It combines things that usually live in separate apps: live meeting transcription, dictation and automated follow-up. Behind it is the company's new speech recognition model, which transcribes in real time across a hundred and two languages. It works for online meetings and for people around a real table, and attendees must be notified that they're being recorded.
Samantha: The interesting part comes after the meeting. ElevenAI lets you search past conversations and ask questions of your meeting archive, say, what did we decide about the budget last month? It can then trigger follow-up actions through ElevenLabs' own AI agent, such as a list of action items or a draft follow-up email. ElevenLabs, which is valued at twenty-two billion dollars, wants this to become one voice-based knowledge hub that replaces standalone note-taking subscriptions.
Bob: What you can do with it: if your team lives in meetings, ElevenAI is worth a look on the ElevenLabs website. Keep in mind it's labelled experimental, so expect rough edges and changes. The announcement doesn't spell out pricing or which devices it runs on, so check those details before rolling it out to a whole team. A sensible first test is dictation on your own, or a short internal meeting, to see how it handles your language and accents.
Samantha: For European teams there's a condition. Under EU privacy and ePrivacy rules, businesses must make sure participants explicitly consent before meetings are transcribed, and a notification is not automatically the same as consent. So before switching it on for internal meetings, agree it clearly with everyone involved and check with whoever handles data protection in your organisation. Meeting archives quickly fill up with sensitive discussions, so it pays to know where that data ends up.
SpaceX's Mechazilla tower catches a Starship spacecraft for the first time, and the footage is online.
Bob: The second one is pure spectacle with real consequences: the footage of SpaceX catching its Starship spacecraft with a launch tower. During Starship Flight 15, the tower SpaceX calls Mechazilla grabbed the upper stage with its giant mechanical arms, nicknamed chopsticks, after the ship came back through the atmosphere from orbit. SpaceX had caught its Super Heavy booster before, but this is the first time the spacecraft itself was caught at the tower instead of splashing down in the ocean.
Samantha: What makes this matter is reuse. SpaceX's plan is that both the booster and the spacecraft return to the launch site, get inspected, refuelled and launched again, closer to an aircraft at a gate than a rocket lost at sea. Catching the ship at the tower is a key step, because it skips an ocean recovery. SpaceX's goal is to drastically cut the cost of transport to other planets. One successful catch validates the concept; it doesn't yet make it routine.
Bob: If you want to see it, Engadget's report on the first chopstick catch of Starship comes with the footage. It's free, takes a few minutes, and works on any phone or laptop. If you've seen the earlier booster catches, this is the comparison worth making: same tower, same arms, but now it's the ship coming home, the part that until this flight had always ended up in the ocean.
Samantha: For Europe, the bigger story behind the video is cost. European satellite operators and space agencies stand to benefit from lower launch prices if fully reusable heavy rockets reach regular operation, because cheaper launches make more missions affordable. That's still an if: this is one flight, and the system is nearing operational readiness, not there yet. But it's the clearest sign so far that the whole rocket coming back could become normal.
Samantha: Today we covered: ChatGPT Wallet letting AI agents shop for you, LangChain's three-billion-dollar valuation, Claude for Manufacturing in factory software, Webb's ice older than its own star, and two things to try yourself: ElevenAI from ElevenLabs, and Engadget's footage of the Mechazilla Starship catch.
Bob: Want to know more or react? Visit stateoftech.eu or email us at info@doorzetters.net.
Bob: State of Tech, the tech world in 15 minutes.