In partnership with
In Today’s Issue:
🧬 An algorithm picked this cancer vaccine, and it worked
🛠️ Five years of work, done in a week and a half
💳 Stripe tells investors the singularity already started
📈 Codex hits 20 million weekly users, and an IPO date
💊 The Ozempic effect that has nothing to do with weight
✨ And more AI goodness…
⚡ The Signal
The biggest AI result of the week arrived as an oncology trial readout.
Merck and Moderna said their individualized cancer therapy met both endpoints in a Phase 3 melanoma study, the first time an mRNA cancer therapy has done that. The AI part is in how the drug is made: a set of AI models reads the sequencing of a patient's own tumor and picks up to 34 targets to encode into the shot, one bespoke product per person. Moderna's stock nearly tripled in a single session. OpenAI's day was all business by comparison, with a customer study in which four Codex agents finished a five-year migration in a week and a half for about $12,000, and a CFO reporting 20 million weekly users for Codex on the way to a 2027 listing. And Stripe, of all companies, told its investors that the singularity started on 1 January and that it has been running the business on that basis ever since. When the company that processes everyone else's payments starts putting a date on it, the argument has left the labs.
All the best,
Kim Isenberg
(OpenAI Codex, via TechCrunch)
🛠️ Five Years of Work in a Week and a Half
A migration Asana's own engineers had scoped at five years took OpenAI's Codex agents about a week and a half. The job was ripping out Enzyme, an obsolete React testing framework that blocked every attempt to upgrade the company's frontend and that nobody had been willing to touch, because the estimate ran to five years of engineering and roughly $6 million in staffing. Up to four Codex agents worked in parallel on separate copies of the codebase from a five-sentence prompt, with engineers reviewing their proposals twice a day. The whole thing came in at roughly $12,000 in model and infrastructure costs.
👉 tl;dr: Five years of projected work, finished in a week and a half. The 500-fold drop in cost is only the second-most surprising number here.
(Stripe CEO Patrick Collison. Getty Images, via TechCrunch)
💳 Stripe Tells Investors the Singularity Already Started
A payments company has told its investors that the singularity began on 1 January, and that it has been running the business on that assumption ever since. In a letter obtained by Axios, Patrick Collison, John Collison and William Gaybrick wrote: "It's a fuzzy and perhaps already overworked term, but we decided that January 1st marked the beginning of the singularity, and we have since been operating on that basis." What convinced them was business data, not philosophy: a sharp break in long-run trends, above all a jump in how many new companies are being founded, and first-half revenue up 41% year over year, which they say the singularity "appears to be accelerating." Collison first used the word half-jokingly at Stripe's Sessions conference in April. Now it is in a letter to investors, next to a $7.5 billion purchase of OpenRouter. TechCrunch thinks the deal itself is far more mundane than the framing: a move into AI expense management.
👉 tl;dr: A lab founder saying we are in the singularity is marketing. The company that clears the payments saying it, and putting a date on it, is a data point.
(OpenAI CEO Sam Altman. Getty Images, via TechCrunch)
📈 Codex Hits 20 Million Weekly Users
Sarah Friar gave an all-hands two numbers on Wednesday, and the one to notice is this: OpenAI's AI coding and work product, Codex, has hit 20 million weekly active users. Agentic coding has moved from developer curiosity to a mass-market tool, and it is the same product Asana just pointed at a five-year migration. Friar also dated the listing: OpenAI "will be a public company in 2027," or sooner if "our business continues to inflect." On Anthropic possibly going public first, she said: "That's OK, we are running our own race." OpenAI filed confidentially in June at an $852 billion valuation, with its revenue run rate up 35% quarter to date and enterprise up 50%.
👉 tl;dr: Twenty million people a week now have an AI writing their code. That is the number the 2027 listing will be priced on.
🎬 Watch This
The Information's AI and robotics reporter Rocket Drew walks through the reporting behind the training pause we covered yesterday. The trigger was an unreleased OpenAI model called Astra, which crossed the company's internal threshold for cybersecurity capability. Drew goes through what OpenAI is actually seeing in that model, and what the monitoring regime it has now committed to is costing.]
"So many breakthroughs are coming."
– Elon Musk, CEO of Tesla, SpaceX and xAI, quote-posting on X, 20 August 2026
He was replying to Nikita Bier, the former head of product at X, who had written: "With cancer vaccines now being discovered with AI ($MRNA), it seems that the US government might actually grow its way out of its budget deficit." Bier is not overselling the AI part. Moderna's own description of the therapy has AI models reading the tumor sequencing and choosing which mutations to go after.
Anthropic has been taking steady criticism over how much Claude Code says before it says anything useful, and overnight it shipped the fix: a built-in Concise output style, set in /config or settings.json. Codex is competing for exactly those developers, which makes the timing look reactive.
An Algorithm Picked This Cancer Vaccine, and It Worked
The Takeaway
👉 Merck and Moderna said on 19 August that intismeran autogene plus Keytruda met both its primary and its key secondary endpoint in the Phase 3 INTerpath-001 melanoma trial.
👉 The companies call it the first positive Phase 3 readout for an individualized neoantigen therapy and for any mRNA-based cancer therapy.
👉 Every dose is bespoke: Moderna's AI models read the patient's own tumor sequencing and pick up to 34 neoantigens to encode into the mRNA.
👉 The Phase 2b version of the same combination was still cutting the risk of recurrence or death by 49% at five years. Full Phase 3 numbers come at a medical meeting.
The first Phase 3 win for a cancer therapy built one patient at a time landed on 19 August, and software chose what went into it. Merck and Moderna said intismeran autogene, given alongside Merck's immunotherapy Keytruda, met its primary endpoint of recurrence-free survival and its key secondary endpoint of distant metastasis-free survival in INTerpath-001, a 1,137-patient trial in people whose stage IIB to IV melanoma had been surgically removed. Patients were randomized 2:1 to the combination or to Keytruda alone and treated for about a year. Professor Georgina Long of Melanoma Institute Australia, who led the study, called it "a landmark moment for adjuvant melanoma treatment."
(Moderna, via NBC News)
There is no shared product here. Every dose is manufactured for one person. A sample of the patient's tumor is sequenced alongside their blood, and, in Moderna's own description, "a series of fully integrated AI algorithms" reads those mutations and "predicts up to 34 of those neoantigens that are most likely to elicit an immune response." Those targets are written into synthetic mRNA, which teaches the immune system to recognize that particular tumor. Moderna says the selection model is built to keep improving as clinical and immunogenicity data come back, and it schedules each patient's manufacturing batch through a separate AI system it calls Maestro. For a decade the pitch has been that AI would help design drugs. Here it is doing the choosing, and the trial worked.
(Professor Georgina Long AO, who led INTerpath-001. Melanoma Institute Australia)
The evidence behind it has been building for years. The Phase 2b run of the same combination, KEYNOTE-942, was still cutting the risk of recurrence or death by 49% five years out, and durability like that is rare in cancer. The open question now is reach. Melanoma is the friendliest possible starting point, because its tumors carry unusually many mutations and hand the model plenty to choose from. Merck and Moderna have nine trials running across lung, bladder and kidney cancer, and every dose in every one of them still has to be sequenced, designed and manufactured for a single person.
Why it matters: Individualized neoantigen therapy has been one of the most expensive bets in oncology for a decade, and the whole approach rests on software picking the right targets out of a tumor's mutations. A Phase 3 endpoint is the first hard evidence that the picking works.
Sources:
🔗 https://www.wsj.com/health/pharma/moderna-merck-vaccine-succeeds-in-preventing-melanoma-from-returning-540e9e18?mod=e2tw
🔗 https://www.merck.com/news/merck-and-moderna-announce-phase-3-interpath-001-trial-of-intismeran-autogene-plus-keytruda-met-endpoints-of-recurrence-free-survival-rfs-and-distant-metastasis-free-survival-dmfs-in-patient/
🔗 https://www.modernatx.com/media-center/all-media/blogs/advancing-fight-against-cancer
Cut Lead Review From Hours To Minutes
Sign up for a free trial of Attio, the agentic CRM.
Ask Attio to build a daily workflow that surfaces the deals that need your attention today, like anything with a stage change, a recent reply, or a new signal in the last 24 hours.
Review your pipeline in Claude, synced live from Attio via MCP.
That's it.
The chart: Moderna's share price through August, posted by the Kobeissi Letter. For two weeks the stock does nothing, sitting around $60. Then the melanoma results land on 19 August and it goes almost straight up. The chart's header reads a close of $183.01, a gain of $111.42 on the day, or +176.97%. Kobeissi posted it that evening in after-hours trading, when the move had reached +212% and more than $50 billion had been added to the company's value.
The lesson: One press release nearly tripled a company. Nobody was buying melanoma sales, because there are none yet and there is no approval. They were buying the idea that a cancer vaccine built for one patient at a time actually works, and that Moderna is the company that knows how to make one.
The caveat: Moderna was a $60 stock for a reason. The shares had spent years sliding as the COVID revenue disappeared, which is part of why the jump is so steep. It also means a single announcement is now carrying a large share of the company's value, and that can come back down quickly.
💊 The Ozempic Effect That Has Nothing to Do With Weight
⚡ Bottom line: Semaglutide cut a key inflammation marker by 37.8%, and the drop began weeks before patients lost meaningful weight.
💡 Why it matters: The heart protection from GLP-1 drugs appears to run partly through inflammation, independent of weight loss.
🔎 What it means: A weight-loss drug is turning into a general anti-inflammatory, which is a far larger market.
The question in one sentence: when Wegovy and Ozempic protect people's hearts, is that only because they made them thinner? A new analysis in Circulation, published 18 August, says no. A team led by Jorge Plutzky at Brigham and Women's Hospital went back into SELECT, the 17,604-patient semaglutide trial in people who already had heart disease and carried excess weight but not diabetes, and pulled out hsCRP, a cheap blood test for inflammation.
Semaglutide cut hsCRP by 37.8% at two years. The fall was visible at four and eight weeks, before any meaningful weight loss, and it showed up even in patients who lost no weight at all. It held regardless of LDL cholesterol or statin use. It also tracked with outcomes: patients who started with more inflammation had a higher risk of heart attack, stroke and cardiovascular death, and those whose inflammation fell had a lower one.
(Wegovy 2.4 mg, the dose used in SELECT. NBC News)
Nothing in this analysis itself was done by AI. The link is the company. Novo Nordisk, which ran SELECT, signed a deal with OpenAI in April to use its models across drug discovery and development, and CEO Mike Doustdar said it would let them "analyse datasets at a scale that was previously impossible." Novo also got early access to GPT-Rosalind, OpenAI's model built for biology, on a short list that included Moderna. Both of today's big medical stories come from companies that now run their research with the same AI lab.
(The Wegovy dose-escalation pack. Novo Nordisk, via STAT)
This is what near-term AI in medicine actually looks like: re-reading finished trials until a decade-old drug turns out to do something nobody designed it for. It is one analysis of one trial, and the authors are careful to say inflammation explains part of the benefit rather than all of it. The interesting part is the direction. The drug is doing something to the body beyond making it smaller, and nobody yet knows how far that goes.
