But where are all the AI-designed drugs?
The world we now live in:
A vocal number of commentators who spent the last two years waxing enthusiastic about AI-designed drugs are now publicly wondering, in apparent confusion, where the AI drugs actually are. Or stroking their chins and sagely noting that, of course, the bottleneck was always data, or clinical development, or some other thing they failed to mention back then.1
The lauded founders of a darling biotech suddenly announce they are departing to start a new venture. What actually happened in the boardroom is not announced. However, industry figures and experts praise the timing and strategy of the transition, studiously ignoring the desultory trial results and bleeding cash reserves sitting behind the curtain.
“Anthropic is now a biotech” zombie posts bounce around LinkedIn, bereft of any explanation of what it means or why anyone should care.
Consultancy whitepapers explain that AI will revolutionize drug development, with detailed charts and numbers attached, but no real account of how these were arrived at.
A pharma company or biotech announces some blue sky target (100 drugs in our pipeline, doing $30B in revenue) and the commentariat praises them for their vision and ignores that it’s mere press release puffery, designed to short up the stock price.
Investors and analysts talking up a new drug class or modality with the certainty of people describing something that has already happened, rather than something that might, while ignoring their earlier and equally certain deliberations on the previous new class or modality.
Most of this doesn’t qualify as news in any real sense. It’s speculation dressed up as reporting, commentary about commentary. There isn’t enough genuinely new, true information in biopharma on any given day to fill the volume of posts and newsletters and opinion pieces claiming to deliver it. Necessarily, most of what circulates has to be noise.
Of course, none of this is new, and none of it is unique to biotech. William Goldman, who spent decades writing and doctoring Hollywood scripts, had a blunt aphorism about the movie business: nobody knows anything. Every time someone explained why a film worked, or didn’t, they were making it up after the fact. The reasons given were stories, built to fit the results.
Biopharma commentary runs on the same trick. Something happens, and within a day there is a tidy, authoritative explanation circulating on LinkedIn, in an investor newsletter, in a trade press quote. It’s not that biotech and drug development is utterly random and facts are irrelevant. It’s that the immediate, neat explanations are almost always a construction, put together after the fact to fit an event nobody saw coming2. It’s that press releases aren’t news but an attempt to set a narrative3.
Recent weeks gave a clean example of the narrative machine running in real time. Reports surfaced that AstraZeneca and Bristol Myers Squibb had held preliminary merger talks4. The commentary that followed was instant and confident. Analysts explained why the deal made sense for BMS (patent cliffs on Eliquis and Opdivo, a need to offset revenue erosion) and why it made much less sense for AstraZeneca (a market-leading pipeline, no obvious need for financial engineering) although maybe it did (a US-listing play tied to AstraZeneca’s recent move to the NYSE). Some called it perplexing5. AstraZeneca’s shares fell as much as 7% on the news, one of its worst single-day drops in years. Perversely, BMS rose about 6%.
Some of this may turn out to be approximately right. But nobody commenting had detailed knowledge of the actual negotiations, the actual motivations of the two boards, details of the deal, or whether the deal will happen at all. Within a day, the market had a fully formed narrative, built on a single-sourced vague report of ill-defined discussions held at some point in recent months6.
Or put it this way: you, I or any one of a number of people could have started that rumor, with no basis in reality. So why are we all talking about something that is near-on phlogiston?
A messy, uncertain, multi-causal event is hard to write about and hard to read about. A clean story with a villain, a hero, a trend, a moral is easy to produce and easy to consume. Journalists have deadlines. Analysts have clients who want a view. The thought-leaderati need material for posts. They all need to show that they know things. The system rewards confident takes over “beats me.” Having opinions, loudly and often, gets you attention7. We’re so ready to hear stories and explanations that we hunt them down whether they’re true or not.8
There’s a pattern here worth calling out: the people with the most knowledge tend to say the least. The boards negotiating a merger, the C-suite that knows why a founder really left, the scientist who knows a drug didn’t get enough time in preclinical, they’re under NDA, managing legal or professional exposure, or busy doing the actual work. The people producing the most voluminous, confident commentary are often the ones with the most time on their hands, the ones trying to impress and sell themselves, the ones farthest from the information.
And there’s no real cost to being wrong. Nobody audits a LinkedIn post from eighteen months ago. The people who were bullish on AI drugs in 2024 aren’t posting corrections now, they’ve moved on to a new hot take, and it costs them nothing. An industry with ten-year development runways gives bad predictions a long time to fade away. Where being wrong costs something, in an actual R&D program with your name on the failed trial, people hedge more and say less.
None of this requires anyone to lie or to be malfeasant. We’ve just incentized over-certainty and de-risked being wrong.
“Don’t believe everything you read” was true, is still true, tempting, and not very useful. There are better questions to ask.
First, in the face of such a volume of low-value information, scepticism is a useful initial stance. “How do you know this? What is it based on?” Very few of the most vocal commentators show their work, instead just making bald assertions.
Second: how fast, and how confident, is this explanation? A true causal account of something genuinely messy - a merger, a departure, a trial failure - usually takes months to surface, if it surfaces at all. A confident explanation that shows up fully formed within a single news cycle was likely assembled to fit the news rather than to explain the news. Plausibility is cheap and plentiful.
I frequently recommend Neil Postman, because he was prescient as regards the modern media and information landscape. My final point comes from him: if this is true, so what? Would it change anything you actually do?9 Most commentary fails this test even when it’s accurate. The founder’s exit story, the consultant’s whitepaper, the conference paper claiming a small increase in accuracy on a limited dataset, none of it would change a single decision any of us make, whether or not it’s true.
There are exceptions, and it’s worth being honest about them. Merger rumors like AZ-BMS are exactly the kind of information that would change behavior for a certain group of people (investors) if they were reliable. Sometimes we do have technical innovations that change the whole field (e.g. AlphaFold). Which is exactly why they get produced and consumed so eagerly. Perhaps the better version of this second question isn’t just “would this change what I do?” but instead “if I acted on this and it turned out to be manufactured, what would that cost me?” This framing separates the potentially useful from the perpetually white noise.
Anthropic is now a biotech? So what. China biotech is getting stronger? Interesting but so what. AZ is merging with BMS? Huh, so what. A big pharma executive said that their data mesh, digital strategy and AI systems are going to accelerate drug development? So what, they all say that. GSK is moving to Cambridge?10 It’ll take years if it ever happens, and even if, so what11. Why did that drug trial fail? We really don’t understand the biology; anything could have happened, so what. Eli Lilly and NVIDIA have announced a co-innovation lab? The spend is less than 1% of their operating cash flow, this is a minor project, so what. Those founders that left? Had made a huge amount of money and failed to deliver; they were probably pushed out, so what. A pharma company has announced a strategic partnership with an AI company? This happens all the time, so what, did they say exactly what the partnership would do …