Alex Telford - Unlocking Innovation in Pharma - [Invest Like the Best, EP.360]
What this covers
The episode is a conversation between the host of Invest Like the Best and Alex Telford, founder of the drug-development software company Convoke. It begins with history. Paul Janssen sets up a Belgian lab in 1952 to modify existing compounds through rational chemistry. Alex then describes two overlapping curves: small-molecule chemistry running out of easy targets, and recombinant DNA opening the way to engineered proteins such as humanized insulin and monoclonal antibodies. He also covers how big pharma has handed most research to venture-funded biotechs and most trials to contract research organizations.
The middle of the conversation concerns evidence. Alex explains why the randomized controlled trial is so valuable, using vitamin D and push-up ability to show confounding by health status. He then turns to surrogate endpoints such as tumor shrinkage, which can speed decisions but can also mislead. The accelerated approval pathway is traced to AIDS activism and early drugs like AZT. The more contested claims are that lower evidence bars pull investment toward cancer and rare disease, and that regulators should publish disease-specific evidence standards in advance.
Economics and speculation fill the rest. Alex describes drug development as a lottery in which a few blockbusters fund many failures, and he argues that pushing hard on prices could starve the whole system. He contrasts the US's largely free pricing with other countries' negotiated prices, and notes how hard CAR T manufacturing is to scale. He also describes how the most impactful drugs come from scientists tinkering against management's wishes, and he uses a Super Mario speedrun simulation to illustrate automated exploration.
Telford argues that pharma's productivity decline is driven less by molecule design than by slow, expensive clinical validation and lottery-like incentives, so the highest-leverage fixes are faster testing, tiered regulatory evidence standards, and protecting bottom-up scientific tinkering.
- Drug development takes about 20 years from target idea to approved drug, and clinical validation rather than molecular design is the binding constraint
- Blockbuster economics work like venture capital, so only a few drugs pay for the failures and small markets go unserved
- Regulators should formalize different evidence standards by disease and enforce post-market data collection
This asset isn't compiled yet
You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.
The accelerated approval pathway grew out of FDA speedups for AIDS drugs such as AZT and ddI during the epidemic, after activists argued that patients facing death should be allowed to try promising drugs, and those early drugs were not especially effective but established a pathway.
“This initial speeding of the regulatory process during the AIDS epidemic turned into what became the accelerated approval pathway.”
The pharma industry operates like a venture capital or lottery model: most drugs fail to recoup costs, and a small number of blockbusters generate enough revenue to pay for all the failures many times over, which distorts incentives so big pharma only pursues drugs with blockbuster potential.
“the way the economics operate is pretty analogous to venture capital and these long-tail models where you have many losers who don't make much money”
Low vitamin D correlates with worse health outcomes in observational data, but trials showed minimal effect from supplementation because vitamin D is largely a marker of poor health, since ill people go outside less and get less sun exposure.
“vitamin D is actually a marker for poor health in general. So vitamin D is produced by your body in response to sunlight. And if you are ill or in poor health, generally, you're less likely to go outside”
Pharma's iteration speed is extremely low compared with tech, where a product demo can be built in days, tested with customers, and iterated near-instantly in code, whereas pharma takes around 10 years to reach clinical trials and weeks to months for animal experiments, so it learns very slowly.
“One of the big problems with pharmaceuticals as an industry is that the iteration speed is super low compared to something like tech.”
A software developer built a tool called Scattershot that simulated millions of Mario jumps across every configuration of jumps and starting positions, finding a much faster way to reach the top of the castle and changing how speedrunning is approached in that game.
“until a software developer developed this tool called Scattershot, which simulates millions and millions of Marios jumping up this level”
The idea that drug discovery is rate-limited by the ability to design molecules is not really true; the actual rate limit is downstream clinical development and testing, including gathering the evidence needed to justify approval and marketing.
“There's this idea that drug discovery is rate limited by our ability to design molecules and design new drugs. And that's not really true.”
From the moment a scientist identifies a disease target to the point a drug is approved, the process takes about 20 years, plus or minus 10, and this timeline has been remarkably consistent.
“as soon as you have an idea for a target, it takes something like 20 years for that to turn into a drug. The sort of time period, 20 years, plus/minus 10, is fairly consistent”
A surrogate is a biomarker such as tumor shrinkage that predicts clinical benefit, and validated surrogates let developers decide to continue or discontinue a drug in years rather than waiting five to ten years for survival outcomes, which is why government agencies should invest heavily in finding them.
“The idea of a surrogate is that you have a measure like tumor shrinkage that predicts benefit down the line, and that helps you develop a drug much faster.”
Since the 2000s big pharma has externalized most research to venture-funded biotechs and clinical trials to contract research organizations, leaving big pharma as commercialization machines whose remaining labs mainly validate external opportunities, and this could extend to commercial analysis and eventually to IP-holding shells.
“So pharma companies -- big pharma companies do a lot of research in-house. They've mostly externalized that to smaller biotechs that are venture capital-funded.”
CAR T-cell therapy requires extracting a patient's immune cells, flying them to a manufacturing site, gene-editing them, and flying them back within a short window because cell viability declines and the patients are gravely ill, which makes the process very hard to scale beyond niche use cases.
“You're taking these cells out of a human, you're flying them to another country probably, you're gene editing them and then you're flying them back.”
Regulators should formalize the principle that different diseases require different standards of evidence and publish in advance the evidence required for each disease, with diabetes requiring large mortality-tested trials because good drugs already exist, while diseases without effective standard care should accept smaller trials and weaker evidence.
“I would want to formalize the idea that different diseases require different standards of evidence and then prepublishing the degree of evidence that we required for each different disease.”
Putting heavy pressure on drug prices can remove the incentives to invest in the fragile pharma ecosystem, because cutting down the few large winners removes funding for the whole portfolio, so society trades off innovation against drug spending.
“If you put pressure too much on drug prices, you can very easily remove the incentives for people to invest in the industry.”
Because cancers and rare genetic diseases have high unmet need and lower evidence requirements for accelerated approval, companies invest disproportionately there, bypassing large Phase III trials that are needed in conditions with established standards of care like diabetes.
“So now what you've seen in the industry is regulatory arbitrage, where a lot of companies are investing in things that are easier to get past the regulators.”
A drug can shrink a tumor without extending patient survival, exposing patients to toxicity while providing no benefit, so tumor shrinkage is an imperfect surrogate that can actually do net harm.
“Sometimes you shrink the tumor, but it doesn't actually help the patient to survive longer in the end. So you're exposing them to toxicity, and you're not helping them survive.”
Being able to do fewer than 10 push-ups predicts higher mortality not because push-ups cause death risk, but because push-up ability is a proxy marker for general health, just as vitamin D levels are.
“It's like with push-ups. They say if you can do less than 10 push-ups, then you're much more likely to die.”
Because R&D spending of hundreds of millions of dollars is paid years before revenue arrives roughly ten years later, companies need very large future revenues in discounted terms for R&D to be worthwhile.
“When you're doing R&D, you're paying money that you have now for revenues in 10 years.”
Paul Janssen, a Belgian doctor whose family imported organ extracts, was motivated by ridicule in medical school to improve on natural compounds; after visiting the US in 1948 he started his own lab in Belgium in 1952 to adapt existing compounds through rational chemistry.
“So Janssen was a Belgian doctor. He trained in medicine, and his dad was actually one of these importers of some of these natural products”
Two curves overlapped in pharma history: the exhaustion of low-hanging fruit in small-molecule chemistry and the growth of biotechnology built on recombinant DNA, which allowed bacteria to produce arbitrary proteins such as humanized insulin and monoclonal antibodies.
“you have sort of two curves overlapping: the falloff of -- the kind of exhaustion of low-hanging fruits in traditional pharmaceuticals and then the growth in what we can do with biotechnology tools”
Much of commercializing a drug is turning a scientific idea into a product that is manufacturable, scalable, and free of unwanted side effects, and many of these problems cannot be discovered until the drug is tested in humans.
“many of these things you don't know about until you've tried it out in humans”
Randomized controlled trials are the most effective way to eliminate biases that investigators cannot foresee, because non-random assignment lets doctors give drugs to sicker or healthier patients, while randomization removes selection bias including hidden sources that cannot be controlled for.
“RCTs are just the most effective way we know of eliminating all these biases that you don't know exist from the start”
Regulators face a near-impossible balance between speed and safety: overly strict rules create an invisible graveyard of drugs that were never developed, while lenient rules let harmful drugs reach market.
“there's going to be a massive, invisible graveyard of people who could have been saved”
The US is mostly a free pricing system where companies charge what the market will bear, while other countries have governments negotiate prices using metrics such as quality-adjusted life years, and both systems have flaws with no perfect solution.
“mostly a free pricing system where you can charge whatever you want.”
The 1950s offered abundant opportunity because molecular tools were new and many starting compounds such as morphine, pethidine, and atropine could be modified, but over time the low-hanging fruit of small-molecule chemistry was largely exhausted, causing traditional pharma productivity to decline.
“the kind of low-hanging fruits have, to some extent, been exhausted with a lot of those molecular tools were developed in that period of time”
In diseases with very few patients, single-arm trials compared against matched natural history cohorts or digital twins can serve as the control, because running a randomized trial with a handful of patients is statistically impossible.
“you compare the progression of a disease with a matched digital twin of these patients, you can train with information from natural history data”
The most impactful drugs usually come from bottom-up, passionate scientists who tinker for years against management objections and secure resources covertly, and management then claims it knew all along, because the impactful drugs cannot be prospectively identified.
“The really important impactful drugs get discovered from this bottom-up process where you have scientists who are really passionate about an idea”
For diseases with too few patients or very slow progression to test in standard trials, regulators should approve drugs based on mechanistic reasoning, animal data, early human testing, and preliminary biomarker evidence, then collect more rigorous data after market launch.
“based on some principal reasons, evidence we've collected about how the drug works in animals and maybe some initial human testing and some preliminary biomarker evidence, we can then approve this drug”
AlphaFold predicts protein structure from sequence, but it is not fully accurate for granular binding-site configurations, and improving it requires more data, which is slow to gather because it was trained on decades of painstakingly collected structures.
“AlphaFold has problems with the very granular predictions of -- the configuration of sites where drugs bind. So it's not completely accurate.”
Radiopharmaceuticals pair a cancer-targeting element with a radionuclide, and because some have a half-life around seven days, they must be manufactured and delivered within a few days before they lose efficacy, making them hard to deliver.
“the half-life of some of these radioactive compounds is something like seven days. You need to manufacture and deliver it to the patient within a few days”
Novo Nordisk was regarded as a boring European insulin company until it became roughly the 15th largest company in the world on the strength of GLP-1 drugs, which came from decades of scientists pushing a long-standing interest in the area.
“Novo Nordisk was regarded as a pretty boring European pharma company, not very innovative, just doing insulin.”
CRISPR is already becoming dated because base editors change specific DNA letters without making double-strand cuts, and prime editing may follow with even more versatile edits, making them more precise than cut-based CRISPR for certain genetic diseases.
“instead of making cuts, you're going to have these base editors that are changing specific letters in the DNA sequence.”
Although drugmakers report roughly 10% to 20% net profit margins and about 80% gross margins, the distortion from R&D timing means most drug companies earn near-zero or very low return on invested capital, making the industry as a whole not a great business.
“actually, most of them are pretty near zero or very low return on invested capital. The industry as a whole is not that great of a business.”
The common US explanation that insurers do not invest in prevention because members switch plans does not fully explain low prevention spending, since nationalized systems like the NHS also underinvest in prevention despite capturing all future benefits.
“if you look at other countries that do have these nationalized health care systems like the NHS, other European countries, they actually don't invest in that much in prevention either”
The US has by far the highest drug prices, about twice European net prices, and accounts for roughly 40% of the global drug market and about 60% of revenue from newer drugs launched in the past decade.
“the U.S. has by far the highest prices in the world. I think people recognize that. It's about twice as high as Europe on a net price basis.”
About 55% of launched drugs follow a path of a brief peak and then slump, failing to earn back even the average development cost, while blockbusters generate 30% to 40% of industry revenue and about 170 blockbuster drugs were generating revenue.
“55% of drugs that launch go through that process. They don't even recoup enough money to pay the average development cost of a drug.”
Using AI and simulation to automate tinkering, as the Scattershot tool did for Super Mario speedruns by simulating millions of jump configurations, could find faster drug candidate paths that humans have not conceived.
“until a software developer developed this tool called Scattershot, which simulates millions and millions of Marios jumping up this level”
Humira's roughly $20 billion in annual revenue looks unreasonable in isolation, but it must be judged in the system of incentives, because the rare huge winners are required to fund the many failures in drug development.
“But you have to think more about the system of incentives.”
A productive use of AI and simulation is to form an intuition about which areas of science are promising and then use automation to do scalable tinkering around that nucleus, tightening the feedback loop so drugs can be developed faster.
“one of the ways you might productively use technologies like artificial intelligence and simulation tooling is to have an intuition about what areas of science might be promising and then use automation to do scalable tinkering”
Many small genetic disease markets with known mechanisms go unserved because a drug earning a few tens of millions a year cannot justify the $100 to $300 million development cost, so reducing trial costs would unlock valuable drugs.
“Even if you get through the whole process of developing this drug, and it works, and you get to clinical trials, you maybe make a few tens of millions a year. And that's just absolutely not worth it because it will cost you $100 million, $200 million, $300 million”
Preventive medicine is underinvested in both the US and nationalized systems like the NHS because the constraint is more on the demand side, with consumers not taking up preventive measures, which leads to a need for a consumer attitude shift before prevention becomes established.
“I feel like the issue is probably more on the demand side where people, unfortunately, don't have a strong demand for preventative medicine”
It is dangerous for an innovation-driven industry like pharma to impose top-down, pre-described views of what to invest in, because the industry is very likely to be wrong about which programs will succeed.
“So I think it's quite dangerous in an industry like pharma, which is so innovation-driven and science-driven, to impose these top-down pre-described notions of what people should invest in. You're very likely to be wrong.”
Overreliance on an accounting-type mindset in pharma is dangerous, because commercial functions should serve R&D, and when managers focused on milking existing products dominate, breakthroughs like Keytruda and GLP-1 agonists become less likely.
“One thing I think is quite dangerous in the pharma industry is just this overreliance on an accounting-type mindset rather than a recognition that it's a science-driven industry”
Complex therapies like CAR Ts and personalized gene therapies require high-touch process expertise and scale that are far beyond traditional pharma's model of making a pill in a factory, and this complexity lets companies build distribution moats.
“So these complex therapies are one way of pharmas trying to build up moats to distribution of products.”
Drugs that extend healthy lifespan will not be developed until researchers identify a surrogate for aging that regulators can accept, because outcome trials of aging would otherwise take too long.
“something like aging, we're never going to get drugs to -- along people's lives until you figure out a surrogate for aging.”
Pharma organizations should hire scientists with good intuitions and allow them to tinker for 10 or 20 years on challenging ideas before commercializing them, since these ideas need that long to mature into products.
“give them enough time, 10 or 20 years, for these ideas to mature into actual products”
After accelerated approval on flimsy evidence, regulators should be far more stringent in forcing companies to run and report confirmatory trials and to remove drugs from market when they fail, because companies drag their heels on confirmatory studies.
“Companies drag their heels on doing the confirmatory study. I think the regulators should be much more stringent in forcing companies to collect that data and report it and take them off the market.”
Alex worries that if innovation incentives weaken, companies will stop investing in fundamental, decade-long research, instead marketing old drugs, extending patents, and investing in hard-to-copy products like radiotherapeutics and CAR Ts.
“I'm actually just going to market the drugs I already have that are old. I'm going to invest in drugs that are very hard to copy.”
The development of many distinct biotech treatment modalities, including CRISPR, RNA interference, CAR T-cells, and base editing, is Alex's most exciting development, because humans increasingly control biological systems with precise molecular tools.
“I think what I'm most excited about is not a specific technology, but more this Cambrian explosion, if you like, of modalities in biotech.”
To predict which technologies will be impactful in the next 10 to 20 years, one should look at what is just nascent today and newly getting approved, such as gene therapies, because one long-term success lets developers iterate the processes that later succeeding products depend on.
“you need to look at what is just nascent today and getting approved. So things like gene therapies”
Alex is short-term pessimistic and long-term optimistic about AI for drug discovery, and he expects AI to help in the near term with downstream tasks such as regulatory documents, market analysis, and trial enrollment rather than with molecular design itself.
“I think I'm short-term pessimistic, long-term optimistic when it comes to AI for drug discovery. And I'm much more optimistic in the short term about using AI to automate some of the processes along with drug development”
Bristol Myers Squibb bought Celgene, a pioneer in CAR T-cell therapy, and has roughly the same number of employees delivering these compounds as the number of patients being treated, around 4,000 each.
“They bought a company called Celgene, who are pioneers in this type of CAR T-cell therapy. And they have the same amount of employees working on delivering these compounds as they have treated patients, so something like 4,000 each.”
Over the next 10 to 20 years, the processes for developing and manufacturing gene therapies will mature and the modality will be applied to many more conditions at larger scale, after the kinks are worked out.
“I think another 10 years, another 20 years, we'll see just the process for developing, manufacturing gene therapies, all those kinks get worked out”
Before the AIDS-era changes, the FDA lacked a process to account for differences in unmet need between conditions, so it would treat diabetes and AIDS similarly, which is why it had to build differentiated pathways.
“And the FDA, until that time, didn't really have a process to deal with these differences in need between conditions. So they would treat diabetes similar to how they would treat AIDS.”
The antibiotic golden age between the two world wars was the pharmaceutical industry's first great success, and many antibiotics were found through experiments that collected soil samples and purified compounds from bacteria, an approach that worked well but was largely limited to infectious diseases.
“a lot of these antibiotics were found through just experiments, going into the wild, find soil samples”
Pharmaceutical companies emerged from apothecaries selling plant and organ extracts of unknown efficacy, and from dye manufacturers who learned to adapt dye chemicals for therapeutic use, with the industry consolidating into its modern form in the early 20th century.
“they were selling extracts of plants”
If doctors choose which patients receive an experimental drug, they tend to give it to sicker patients who need it, which makes the drug group look worse even if it works, so non-random allocation creates selection bias.
“Then the doctors may be more inclined to give the drug to patients who are sicker who really need it.”
Early pharmaceutical companies sold natural extracts such as plant and organ extracts, including thymus gland preparations, even though nobody knew which parts actually treated disease, so most of these products did not work in most cases.
“we don't really know what parts of the natural extracts actually treat health”
Monoclonal antibodies are now the top-selling drug class, with Keytruda and previously Humira as the leading products, and the processes to develop, produce and sell them at scale are well established, though it took a long time to reach that point.
“monoclonals now are the top-selling drug in the world, Keytruda. Previously, it was Humira. They're both monoclonal antibodies.”
Zolgensma treats spinal muscular atrophy, a devastating infant neuromuscular disease where the most severe type previously killed patients before age one, and it can now seemingly almost cure patients, though gene therapies have had few other commercial successes so far.
“Patients that have that disease would die before they are one year old in the most severe type. But now they could be seemingly almost cured with these gene therapies”
RCTs should remain the gold standard and be used whenever feasible, with alternatives such as single-arm trials and digital twins reserved as tools where RCTs are infeasible.
“So I think RCTs are always going to be the gold standard.”
People would rather take a drug like Ozempic to treat obesity than sustain diet and exercise, which shows a consumer attitude that undermines the demand for preventive medicine.
“So they would rather just take Ozempic when they are obese and have that treat their obesity.”
Drug discovery starts in academia, largely funded by government research such as the NIH, where scientists map the basic mechanisms of diseases and identify genes and proteins that could be modulated to treat them, but this is only a small part of the story because turning those targets into drugs requires mostly for-profit biotech work.
“So it starts with academic research. This is mostly funded by government research and it occurs in universities. So the NIH is the biggest funder of biomedical research in the world”
Healthy people enrolled in plans pay most of the money in the system, so they should want the system to provide value for money when they become sick, and their demand for preventive products should incentivize companies and insurers to provide them.
“it's actually healthy people who are enrolling in the plans and are paying the majority of the money that goes into the system”
Externalization has trade-offs and creates redundancy, because biotechs preparing to launch must build up duplicate capacity to run trials efficiently, even though the trend has some benefits.
“externalization hasn't been completely helpful. There's been some trade-offs there. But there's a lot of redundancy in the industry.”
The first CRISPR-based therapy has been approved in both the UK and the US.
“we just got our first CRISPR-approved therapy in the U.K. and the U.S.”
Alex said the kindest thing anyone has done for him is his parents' unending support, which helped him recover from many low points in his life.
“my parents have always been -- unending well of support, and they build me out of many low points in my life.”