Mark Delaney: Juniper, good week? Because mine got weird the second I read this Reuters headline and I've been kind of chewing on it since.
Juniper Vale: The Insilico thing? I read it twice. What got you?
Mark Delaney: The gap. Uh — so here's the headline claim: Alex Zhavoronkov, CEO of Insilico Medicine, tells Reuters that their AI platform, Pharma.AI, took a drug candidate from scratch to preclinical in nine months. Four and a half years is the industry standard. That's the win they're announcing.
Juniper Vale: Okay, and you're already squinting at something.
Mark Delaney: Their actual drug, Rentosertib — their first AI-designed candidate — is sitting in Phase II trials right now. Phase II. And no AI-discovered drug has cleared regulators anywhere on earth yet. So we're celebrating the sprint while the marathon hasn't finished.
Juniper Vale: Wait — zero approvals? I want to name that clearly because I think people are going to assume otherwise.
Mark Delaney: Zero. And that's kind of the whole episode, isn't it? What did Insilico Medicine actually prove, and what are we just… hoping is true?
Juniper Vale: That's exactly the question — and I think the word preclinical is where we have to start, because that word is carrying so much weight in Zhavoronkov's announcement and most people are going to slide right past it.
Mark Delaney: Yeah, and I think people hear 'preclinical candidate' and they're like, okay, so it's almost a drug. Like it's in the waiting room.
Juniper Vale: Think of it like this — drug development is building a house. Preclinical is the architect handing you a blueprint. That's what Insilico did in nine months. The blueprint is real, it's a genuine step, but nobody's living in the house yet.
Mark Delaney: And most blueprints never become houses.
Juniper Vale: Right — that's the number that landed hard for me. Clinical success rates historically are under ten percent overall. Meaning nine out of ten drug candidates that look great on paper never make it to approval. And for AI-designed drugs specifically? That number is, uh — I mean, it's unknown. There's no track record yet.
Mark Delaney: Wait — under ten percent across the whole industry, and for AI it's just… blank? No data at all?
Juniper Vale: Completely unestablished. Rentosertib in Phase II is the closest data point we have, and the answer isn't in yet. So what Pharma.AI actually proved is that generative AI can compress the hypothesis-generation phase — the part where chemists are manually designing and testing molecules. Insilico was synthesizing somewhere between 60 and 200 molecules per program instead of thousands. That's genuinely new. But that's discovery. The clinical proof is a different building entirely.
Mark Delaney: So the nine months — that's only the first, what, ten to twenty percent of the whole development timeline?
Juniper Vale: Discovery is roughly ten to twenty percent of the total timeline. Making it faster is real — I don't want to dismiss it — but the eighty to ninety percent that's left? Clinical trials. That clock hasn't changed. The patient waiting on Rentosertib is still looking at years.
Mark Delaney: But that's actually where I want to push on the coverage, because the thing I keep seeing is — everyone's writing this as a pure AI story. Software breakthrough. Algorithm wins. And I'm not sure that holds up when you look at how Insilico actually runs.
Juniper Vale: What do you mean by how they run?
Mark Delaney: Their model is split — uh, frontier AI research sits in Montreal and Abu Dhabi, but the actual experimental validation, the lab work, the scaling? That's China. And China isn't just a location. It's lower costs, faster lab throughput, regulatory friction that moves quicker than what you'd hit in Basel or Boston.
Juniper Vale: Okay, I want to stress-test that a little. The 60-to-200 molecules per program — that efficiency is real regardless of which country the lab is in, isn't it? Like, that's the algorithm doing fewer wrong guesses upfront.
Mark Delaney: No, that's fair — I'll give you that. The molecule count is genuinely lower and that's the AI doing work. But you can't actually separate nine months from the infrastructure that made nine months physically possible. Take the same algorithm, put it in a Western lab with Western costs and Western regulatory checkpoints — does it still run in nine months? I'd want to see that before calling it a software breakthrough.
Juniper Vale: And here's what makes this harder to untangle — Zhavoronkov himself told Reuters that most Insilico programs average 12 to 18 months. Not nine. So the nine months is the record, not the routine.
Mark Delaney: Wait — he said that? To Reuters?
Juniper Vale: That's the number he reported, yeah. Which means the headline they're leading with is the outlier. And I'd add — this is all self-reported. Reuters published it, but it hasn't been independently verified, so we're hedging here too.
Mark Delaney: And honestly, the only thing that actually settles any of this — the AI question, the infrastructure question, the nine-months-means-something question — is what happens when Rentosertib's Phase II readout comes in. That's the one number that makes the whole record either real or a very expensive footnote.
Juniper Vale: And that's the whole weight of it, right — the Phase II readout isn't just Rentosertib's report card. It's retroactively scoring the nine-month record. Picture a patient with a rare autoimmune condition, sitting in a waiting room in 2024, they've got a Reuters alert on their phone saying 'AI designs drug in 9 months,' and they think something is almost ready for them. What's actually ahead? Phase II, then Phase III, then regulatory review. That's four to six years minimum — and that's the optimistic version where it works.
Mark Delaney: Four to six years after the nine months they already heard about.
Juniper Vale: The nine months didn't touch that patient's timeline at all. That's what actually happened.
Mark Delaney: Okay but — and I don't want to just be the cynical guy here — there is a real bull case, isn't there? Like, even if Rentosertib's clinical outcome is the same as any traditionally discovered drug, faster failure frees up capital. You run more programs in parallel. The speed compounds over iterations, not just one molecule.
Juniper Vale: That's actually — yeah, that's the best case for it. But it only holds if Insilico reinvests. If the nine-month record becomes a press cycle and not a new program starting the next morning, the compounding never kicks in.
Mark Delaney: And there's another layer I don't think Zhavoronkov touched in the Reuters announcement at all — what happens when these drugs go to Western regulators? Like, the data trail for Rentosertib runs through Chinese labs. Is the FDA just... going to take that at face value?
Juniper Vale: I mean, that's genuinely unaddressed. The geopolitical friction is real — a regulator in the U.S. or Europe seeing China-validated preclinical data could demand additional scrutiny, maybe replication studies, and that adds time back to the very pipeline Insilico is claiming to have shortened.
Mark Delaney: So the speed advantage gets taxed at the border, kind of.
Juniper Vale: That's a pretty clean way to put it. Which is why — you know — the Rentosertib Phase II readout is the only number I'd watch. Not the nine months, not the molecule count. If it shows superior efficacy compared to traditionally discovered drugs, the record earns its meaning retroactively. If it fails, or just performs the same, the speed advantage was real and also completely beside the point.
Mark Delaney: Alex Zhavoronkov and Insilico Medicine did something real. The nine months, the molecule efficiency, the Pharma.AI platform doing work that used to take years of manual chemistry. That's not nothing. But I can't get past this — what does it mean to call something a breakthrough before a single patient has actually been helped by it? Is the record the achievement, or is the record just... the setup for the actual achievement?
Juniper Vale: I don't have a clean answer to that. I genuinely don't.
Mark Delaney: Yeah. Neither do I. And I think — uh, I think that's the honest place to stop. The next inflection point isn't a faster discovery engine. It's the Rentosertib Phase II readout. That number will say more about what the nine months was worth than anything Zhavoronkov told Reuters.
Juniper Vale: Thanks for chewing through it with me. It's a better question than I started with.
