Claude finds an enzyme, ChatGPT Ads hit Asia, Muse goes shopping.

On September 23, Anthropic announced a life sciences research group, formed this spring, with its own laboratory in the Bay Area. Its first headline result: about 950 Claude agents searched DNA data for 21 hours, used 210 million tokens, and turned up a new family of enzyme systems Anthropic calls array-associated reverse transcriptases, or ART. Humans wrote the opening prompt and did the lab work. The agents did the searching in between.

Anthropic admits nobody knows yet what ART does. The company released a preprint early on purpose, and says the work to understand it is ongoing. CRISPR pioneer Feng Zhang, quoted in the post, called the finding "genuinely intriguing and merits further investigation." From a scientist, that is a standing ovation.

Also on September 23, Amazon said a new selling partner plugin launches with Amazon Quick and in beta with Claude. Seller Assistant now remembers each seller's pricing patterns, inventory cycles and growth goals, and that memory follows the seller across Seller Central, Quick and Claude. New always-on workflows can watch conditions and act without the seller logging in, with an audit trail on every action and guardrails the seller sets, from recommendations only to actions with approval.

Amazon's own numbers: the assistant reaches over 90% of its selling partners, has hundreds of thousands of active users, and those users accept its recommendations over 90% of the time. It runs on Amazon Bedrock with Claude models, and Amazon says seller data stays inside its infrastructure. Primary account holders also get a free 12-month Amazon Quick Plus subscription in an offer that runs through December 31, 2026.

So the assistant that knows your restock cycle now also has a chat window with a very good vocabulary.

Marketing insight: If you sell on Amazon, try the Claude beta on one product group before you trust it with the whole catalog. Set it to recommendations only for the first month, write your price floors down as guardrails, and read the audit trail every Friday.

OpenAI started rolling out ChatGPT Ads in Indonesia, Malaysia, the Philippines, Singapore, Thailand, Vietnam and Taiwan on September 23, which puts the product in more than 60 countries. Advertisers can buy through OpenAI's sales team, big agency groups such as dentsu, Havas Media, Omnicom Media, Publicis Groupe and WPP, or self-serve through Ads Manager if the business is eligible. Ads still show only to Free and Go users. Plus, Pro and Enterprise stay ad-free.

OpenAI repeated two numbers from late August: ads hit a $1 billion annualized revenue run rate in under 200 days, and tens of thousands of advertisers are running them. It also said new formats, tools to tune campaigns and measurement are coming over the next few months. So the measurement is still under construction, and so is the competition.

Marketing insight: If you already sell in any of these markets, or anywhere ChatGPT Ads run, put a small test budget on it now, while the auction is young and the tools are thin. Write the ad for the question a buyer would type, not the slogan on your homepage.

At Connect on September 23, Meta announced a stack of Muse updates. Muse now holds real-time voice conversations and keeps working in the background while you talk. Muse for Mac gained computer use, so with permission it can drive any app on the Mac. Muse will get its own email address, and it arrives on Meta's AI glasses in the coming months, where Meta says you can ask about a product on a shelf and Muse can take action.

Then the shopping list. New connectors include Walmart, Best Buy, American Eagle Outfitters, DICK'S Sporting Goods, Fanatics, Gap, Michael Kors, Sephora, Ulta and Wayfair, plus Shop Pay and PayPal for payments and Instacart for groceries, on top of the Shopify catalog access Muse launched with. Expedia for travel is coming soon. The most important shopper this quarter has no thumbs.

In all four stories, a machine reads more than a person could, faster, and then does something about it.

For marketers, complete and checkable now beat clever: clean product data, real customer language, numbers you can source, and a person who verifies what the machine concluded before anyone spends money on it.

If you want help deciding which of these to test first and which to let other people pay to learn, that is the exact conversation our founders like having. Talk to them below. See you in the next edition.