How a Medicine Moves from a Scientific Idea to a Pharmacy

TL;DR. Picking up a prescription takes about three minutes: a name is confirmed, a bottle is scanned, a pharmacist says a sentence about how to take it. That bottle is the visible end of a process that, for most drugs on the market, started ten to fifteen years earlier as an idea about a single molecule in the body, survived lab and animal testing, then three phases of human trials, then a regulatory review running to thousands of pages, then a manufacturing and distribution chain built to standards enforced by unannounced inspections. Roughly nine out of ten drug candidates that reach human trials are abandoned, and the ones that survive cost, on average, somewhere between one and three billion dollars, depending on whose accounting you trust. Almost none of that history is visible in the bottle. Most of it exists because, in 1962, Congress found out what happens when it isn't there.

Key takeaways

  • Filling a prescription looks routine because a pharmacist has already absorbed the risk: verifying it, checking it against a patient's recorded allergies and other medications, and counseling on how to take it are legally required steps, not customer service extras.
  • Most drugs work by binding to one specific biological target, commonly a receptor or an enzyme, though "one drug, one target" is a simplification; many drugs act on several targets at once, and some work through mechanisms unrelated to any receptor.
  • Getting a molecule from a laboratory idea to a pharmacy shelf runs through preclinical testing, three phases of human clinical trials, formal regulatory review, and only then manufacturing and distribution. Roughly 90 percent of drugs that enter human trials never reach approval.
  • The entire modern approval system exists because of a specific historical failure: the 1962 Kefauver-Harris Amendment, passed after the thalidomide birth-defect tragedy, was the first U.S. law requiring proof a drug works, not just that it's safe.
  • Generic drugs, more than 90 percent of U.S. prescriptions filled today, exist as a fast, low-cost approval pathway only because of a 1984 law, the Hatch-Waxman Act, that balanced brand-name patent protection against faster generic entry once patents expire.
  • The system doesn't stop watching once a drug is approved. Manufacturing quality control continues for the product's life, and a federal reporting system keeps tracking side effects that trials are too small and short to catch on their own.

Three minutes at the counter

You give a name, sometimes a birth date. A pharmacy technician pulls up the order, a small orange or white bottle appears from a shelf or a robotic dispensing carousel, and a pharmacist steps over, asks if you have any questions, and says something short: take this with food, this might make you drowsy, don't stop taking it early even if you feel better. You pay a copay and leave. The entire visible transaction takes less time than it took to drive to the pharmacy.

Nothing about that moment suggests what sits behind it. For a genuinely new drug, one that didn't exist as a treatment for anything a decade or more earlier, that bottle is the last stop on a process that began in a laboratory with a hypothesis about a single molecule inside the human body, survived years of tests designed to kill weak candidates before they could hurt anyone, and then passed through a federal agency empowered, since 1962, to say no. Most of the molecules that started down that path did not make it. The one in your hand did, and that is precisely why it's boring to pick up. Boring, here, is the product of decades of work.

What actually happens when a prescription is filled

A pharmacist's job at the counter is smaller in scope than the manufacturing story behind the pill, but almost none of it is optional. Federal Medicaid rules, and state pharmacy laws modeled on them since a 1990 federal law usually called OBRA '90, require three steps every time a prescription is filled: maintain a patient medication record, run a prospective drug utilization review against it before dispensing, and offer to counsel the patient on the drug.

That review happens invisibly, in the seconds between a prescription arriving and a label printing. Pharmacy software checks the new prescription against everything on file for that patient: other prescriptions, recorded allergies, the diagnosis if it's on file. It's looking for therapeutic duplication (two drugs that do the same thing), a dangerous drug-drug or drug-disease interaction, an incorrect dose, or a known allergy. If something flags, the pharmacist has to use professional judgment about whether to fill the prescription as written, call the prescriber, or talk to the patient first. Only after that review clears does counseling happen, and the offer has to be made out loud; a sign saying counseling is available does not satisfy the legal requirement.

Underneath that is a quieter question: what is the pill actually doing once it's swallowed? The honest answer depends on the drug, but a large share of prescription medicines work by binding to one specific biological target, most often a receptor (a protein, usually on a cell's surface, built to recognize one particular molecule and trigger a response when it attaches) or an enzyme (a protein that speeds up a specific chemical reaction). A drug that attaches to a receptor and triggers the same response the body's own signal would is an agonist; one that blocks the response instead is an antagonist. A drug that blocks an enzyme's active site is an inhibitor: statins lower cholesterol this way, and beta blockers treat high blood pressure by blocking a receptor that would otherwise speed up the heart. That's the textbook picture, and it's real, but it isn't universal. Antacids and laxatives work through plain chemistry rather than binding anything, and many real-world drugs act on more than one target at once, sometimes on purpose, sometimes as a side effect nobody fully understands even after approval.

Don't be confused: "FDA approved" does not mean the FDA ran the tests. The Food and Drug Administration does not discover drugs or run clinical trials. A pharmaceutical company, or a university lab that later partners with one, designs and pays for every study. The FDA's job is to review the resulting data and decide whether it meets the legal standard for safety and effectiveness. "FDA approved" means a regulator examined someone else's evidence and agreed it was strong enough, not that government scientists independently verified the drug themselves.

From a target to a molecule

Long before any of that reaches a pharmacy, a drug starts as a guess about biology. Researchers begin with target identification: finding a receptor, enzyme, or other molecule that plays a causal role in a disease, something that, if blocked, activated, or otherwise altered, should change the course of the illness. Validating that guess in cells or animal models is its own multi-year research problem, and a wrong guess here is why many programs fail before a single human trial begins.

Once a target looks real, the search for a molecule that can hit it reliably starts. High-throughput screening runs tens of thousands to millions of candidate compounds against the target using automated robotics, looking for any that bind well; a computational alternative starts instead from a model of the target's shape and designs a molecule to fit it directly. Either way, the initial "hits" are almost never good enough on their own; a hit-to-lead and lead optimization phase chemically tweaks the most promising candidates for years, improving potency, reducing toxicity, and making sure the molecule can survive being swallowed or injected and still reach its target intact.

Whatever survives that gauntlet moves into preclinical testing: laboratory studies in cells and then in at least two animal species, checking toxicity, how the body absorbs and eliminates the compound, and whether it damages DNA or harms reproduction. This stage commonly takes several years and has to follow its own federal quality standard, Good Laboratory Practice, before a company can file an Investigational New Drug (IND) application asking the FDA for permission to test the compound in people. The FDA has 30 days to object; if it doesn't, human testing can begin.

Three phases, thousands of people

Human testing is where a drug candidate meets the population it's meant to help, in three deliberately escalating stages, and it's also where most candidates die.

Phase 1 answers a narrow question: is this safe enough to keep testing, and what dose does the body tolerate? It typically enrolls 20 to 80 people, often healthy volunteers rather than patients, and runs for a year or two. About 70 percent of drug candidates that begin here pass.

Phase 2 shifts the question toward whether the drug actually works, in a few dozen to a few hundred patients who have the condition it's meant to treat, while continuing to watch for side effects and refine dosing. This is historically the deadliest phase: only around a third of drugs that enter Phase 2 make it out, because it's the first point where a biologically plausible idea meets real disease and often doesn't move the needle enough to justify continuing.

Phase 3 is the large confirmatory study, often several hundred to a few thousand patients, sometimes tens of thousands for a common condition, usually compared against an existing treatment or a placebo, run long enough to catch side effects a smaller trial would miss and to generate evidence a regulator will accept. Roughly a quarter to a third of candidates that enter Phase 3 come out with data strong enough to file for approval.

Multiply those numbers through the whole pipeline, from the first Phase 1 dose to an approved drug, and independent analyses of large trial datasets put the overall odds of success at roughly 10 to 14 percent, varying enormously by disease area (infectious-disease vaccines succeed far more often than cancer drugs). Put plainly: across every drug candidate that a real human being agreed to take in a clinical trial, something like nine out of ten never reach a pharmacy shelf. Every one of those trials, including the ones that fail, depends on patients who volunteer with no direct benefit to themselves, and their willingness to enroll is the resource the entire discovery pipeline runs on.

The regulator's desk

Once Phase 3 succeeds, the company compiles everything, every study, every adverse event, every manufacturing detail, into a New Drug Application (NDA), a submission that for a genuinely novel drug can run to hundreds of thousands of pages. FDA reviewers, organized by discipline (medical officers, statisticians, chemists, pharmacologists), examine that record rather than take the company's summary at face value, checking the statistical analysis independently and inspecting the manufacturing sites where the drug will actually be made.

Since 1992, this review has run on a set clock under the Prescription Drug User Fee Act (PDUFA), which lets the FDA charge pharmaceutical companies fees that fund additional review staff, in exchange for committing to review timelines: a standard review targets 10 months from filing, and a priority review, granted to drugs offering a significant improvement over existing treatments, targets 6 months. Before PDUFA, average review times ran 21 to 29 months; the law roughly cut that in half, though the timeline is a performance target the agency usually meets, not a hard legal deadline. If the review succeeds, the FDA approves the drug for specific, labeled uses, and only then can manufacturing at commercial scale and distribution to pharmacies begin.

Turning a molecule into a pill

An approved drug still has to become a physical, swallowable object, and that split is real inside the industry: the chemical or biological substance that produces the therapeutic effect is the active pharmaceutical ingredient (API), manufactured first, often at a dedicated chemical plant, sometimes on a different continent than where the final product ends up. The API alone is rarely what a patient takes; it's combined with excipients, inactive ingredients chosen for reasons that have nothing to do with treating the disease: binders that hold a tablet together, coatings that control how fast it dissolves, fillers that make a microgram-scale dose physically possible to handle. Turning API and excipients into a finished tablet, capsule, or injectable vial is formulation, tested almost as rigorously as the drug itself, since a formulation that dissolves too fast, too slow, or unevenly can change how much API actually reaches the bloodstream even though the chemical identity of the drug hasn't changed at all.

The trip to the pharmacy shelf

Finished product leaves the manufacturing plant and almost never travels directly to a pharmacy. Instead it moves through pharmaceutical wholesalers, companies that buy in bulk from manufacturers and redistribute to individual pharmacies, hospitals, and clinics. In the United States, this layer is remarkably concentrated: three companies, McKesson, Cencora (formerly AmerisourceBergen), and Cardinal Health, together handle more than 90 percent of pharmaceutical distribution revenue, a combined 776 billion dollars in 2024 alone. For a drug that needs to stay refrigerated or frozen along the way, this leg runs through the same kind of temperature-controlled logistics chain covered in how food, medicine, and vaccines stay cold in transit; nothing about that need disappears just because the product started as a laboratory target rather than a vaccine.

Who keeps it running

A prescription bottle carries the fingerprints of an unusually long list of professions. Medicinal chemists and biologists spend years turning a target into a workable molecule, long before a single patient is enrolled. Clinical research coordinators run the day-to-day machinery of a trial site: screening volunteers against strict eligibility rules, drawing blood, tracking every visit. The patients who agree to be studied, healthy volunteers in Phase 1, people living with the disease in Phases 2 and 3, are themselves an essential, largely uncompensated part of the workforce. Biostatisticians design the trial's statistical plan and analyze the results, work that determines whether a real effect gets recognized as real or dismissed as noise. FDA reviewers, organized into disease-specific divisions, examine that evidence independently rather than signing off on a company's summary. Manufacturing quality control staff test samples from every production batch against approved specifications, a job that continues for as long as the drug stays on the market. Pharmacists and technicians do the final, patient-facing check described above. And, less visible but hugely influential in what a prescription costs, pharmacy benefit managers (PBMs) negotiate drug prices and rebates on behalf of insurers and decide which drugs a health plan covers, sitting between manufacturers and pharmacies in ways that shape the U.S. drug market as much as any single law does. Three companies, OptumRx, Express Scripts, and CVS Caremark, handle about 79 percent of prescription claims in the country, a concentration under sustained scrutiny from regulators and Congress.

Where this came from

Federal authority over drug approval arrived in two large jumps, each one following a public catastrophe, not because an agency decided it needed more power.

The first jump was the Pure Food and Drug Act of 1906, passed after Upton Sinclair's novel The Jungle exposed unsanitary conditions in meatpacking and food adulteration more broadly. The 1906 law required that a drug's active ingredients appear on its label and that drugs meet purity standards set by existing pharmacopeias, and it created the federal enforcement body that would eventually become the FDA. It said nothing, however, about whether a drug actually worked. A manufacturer in 1906 could sell a truthfully labeled product that did nothing for the condition it claimed to treat, and the law had no mechanism to stop them.

The second, far larger jump came in 1962, and it is the single most consequential piece of drug regulation in U.S. history. In September 1960, Richardson-Merrell applied to the FDA to sell thalidomide, already marketed in West Germany and dozens of other countries as a sedative and morning-sickness remedy, under the brand name Kevadon. The reviewing officer, Frances Oldham Kelsey, on one of her first assignments at the agency, judged the safety data inadequate and used the FDA's authority to withhold approval in renewable 60-day increments while she kept requesting more information, despite over a year of intense company pressure. Before it broke, doctors in West Germany and Australia identified a wave of severe birth defects, including limbs that failed to develop normally, in babies whose mothers had taken thalidomide during pregnancy. Worldwide, roughly 8,000 to 10,000 infants were affected, with thousands more pregnancies ending before birth. Because Kelsey had never approved the drug for the U.S. market, harm here was limited mostly to physician test samples, but the near miss, and the disaster's scale elsewhere, landed with full force on Congress.

The result was the Kefauver-Harris Amendment, signed by President Kennedy on October 10, 1962, and it changed the legal bar for every drug sold in the United States afterward. For the first time, a manufacturer had to prove not just that a drug was safe, but that it worked, supported by "adequate and well-controlled investigations," the language that underlies the modern three-phase trial system described above. It also gave the FDA authority to pull already-marketed drugs that failed the new standard, expanded oversight of manufacturing, and tightened rules on drug advertising. Kelsey received the President's Award for Distinguished Federal Civilian Service the same year, one of the few times a single reviewer's judgment call is directly traceable to a change in federal law.

Standards that make it work

Approval is a single decision. Keeping a drug identical, batch after batch, year after year, requires an ongoing rulebook: Current Good Manufacturing Practice (CGMP), codified in federal regulations covering facilities, equipment, personnel training, sterilization, and record-keeping at every plant that makes drugs for the U.S. market, domestic or foreign. "Current" is deliberate: a facility must keep updating its methods as technology improves, not simply repeat whatever passed inspection a decade earlier. The FDA enforces this through unannounced or scheduled inspections, typically every two to four years for a routine facility, and a plant that fails can have its drugs declared legally "adulterated," halting shipments regardless of whether any specific batch has been shown unsafe.

The clinical trial success rate mentioned earlier, roughly 10 to 14 percent of candidates entering human trials eventually reaching approval, is itself treated within the industry almost as a standard, the baseline figure investors and researchers use to judge whether a pipeline is performing normally. A cancer program succeeding at 3 to 4 percent isn't a sign something has gone wrong; it's within the historical range for that disease area.

Generic drugs run on a separate, deliberately faster standard, created by the Hatch-Waxman Act of 1984 (formally the Drug Price Competition and Patent Term Restoration Act). Before 1984, a generic manufacturer selling a copy of an off-patent brand-name drug had to repeat much of the original safety and efficacy testing from scratch, an expensive requirement that kept generic competition rare; generics then accounted for only about 19 percent of prescriptions filled. Hatch-Waxman created the Abbreviated New Drug Application (ANDA), letting a generic manufacturer rely on the FDA's existing determination that the brand-name original is safe and effective, provided it demonstrates bioequivalence: delivering the same active ingredient to the bloodstream at essentially the same rate and extent as the brand-name product. In exchange, the law extended patent protection for brand-name manufacturers to partly offset time lost to FDA review. Generics now account for more than 90 percent of all U.S. prescriptions filled.

Don't be confused: bioequivalent does not mean untested. A generic manufacturer does not repeat the original Phase 1 through 3 trials, but it does run its own studies proving the generic releases its active ingredient into the body the same way the original does, and its plant has to pass the same CGMP inspections as any brand-name facility. What's skipped is duplicating trials that already answered whether the ingredient itself works; what isn't skipped is proving the specific generic performs the same way and is made to the same quality standard.

Keeping it working

Approval and initial manufacturing inspection are not the end of the FDA's involvement; they're closer to the midpoint. MedWatch, the FDA's safety reporting program, and the FDA Adverse Event Reporting System (FAERS) that collects the reports, exist because clinical trials, even large Phase 3 studies, are fundamentally limited: they run for a fixed time, on a few thousand selected patients, and cannot catch a side effect that shows up only after a year of use, or only in one patient in fifty thousand. Healthcare professionals can report suspected adverse events voluntarily; manufacturers are legally required to. FDA safety evaluators then look across the accumulating reports for patterns, called safety signals, that no individual doctor could see alone. A confirmed signal can lead to an updated warning label, a letter to prescribing physicians, or, in serious cases, a request that a drug be pulled from the market.

Manufacturing quality control runs on the same permanent basis. Every batch produced, not just the batches used to win initial approval, has to be tested against the specifications on file before release, and CGMP compliance is checked on a recurring inspection cycle for as long as a plant keeps making the drug. Approval is a snapshot of one moment; the manufacturing standard behind it is meant to hold indefinitely.

When it breaks

Thalidomide is the failure that built the modern system, but the system it built is not failure-proof, and two very different kinds of breakdown have recurred since 1962.

The first is what post-market surveillance exists to catch: a real side effect trials were too small or short to detect. Rofecoxib, sold as Vioxx, was approved in 1999 as a pain reliever and was, at its peak, taken by more than 80 million people worldwide. A long-term colon-polyp prevention study, not the original approval trials, found in 2004 that patients taking it for 18 months or longer faced roughly double the risk of heart attack or stroke. Merck withdrew the drug that September, at the time the largest prescription drug withdrawal in history; a later analysis in The Lancet estimated Vioxx use had caused on the order of 88,000 heart attacks in the United States, about 38,000 of them fatal. Warning signs, including a 2001 academic paper and a 2002 Health Canada advisory, predated the withdrawal by years, which is why the case is still cited in debates over how fast a regulator should act on an accumulating but inconclusive signal.

The second kind of failure is quieter and, today, more common: the drug shortage. In late 2022, an Intas Pharmaceuticals plant in Ahmedabad, India, a major supplier of the injectable chemotherapy drugs cisplatin and methotrexate, failed a surprise FDA quality inspection and shut down. Sterile injectable manufacturing is hard to stand up elsewhere, and thin margins had already pushed that market down to a handful of suppliers, so the shutdown alone knocked out roughly half of U.S. cisplatin and methotrexate production. Cisplatin went into nationwide shortage in February 2023; carboplatin followed by April. A survey that fall found 93 percent of U.S. cancer treatment sites reporting a carboplatin shortage and 70 percent reporting one for cisplatin. That's the structural weak point in the generic system: Hatch-Waxman made generics cheap and common, but the low margins that make them cheap also discourage manufacturers from keeping redundant capacity, so one quality failure at one plant can shortage a drug with no patent and plenty of eligible manufacturers on paper.

The scale of it

Estimates of what it costs to bring one new drug to market vary enough to be a running argument in health policy circles, not a settled number. The Tufts Center for the Study of Drug Development, using confidential data from pharmaceutical companies on 106 drugs, put the average full cost, including capital tied up over a decade or more of development, at 2.6 billion dollars, over a typical ten-year timeline. A 2020 study in JAMA, using only public financial disclosures from 63 companies, arrived at a considerably lower median of 985 million dollars and an average of 1.3 billion. Both describe the same reality, that most candidates fail and their costs are absorbed by the minority that succeed, but they disagree sharply on the dollar amount, largely because they draw on different data from different funding sources with different incentives.

Downstream of that spending, the FDA's Center for Drug Evaluation and Research approved 50 novel drugs, new molecules never before marketed in the U.S., in 2024, and 55 in 2023, near the top of the agency's own thirty-year range; the rolling ten-year average now sits at roughly 46 to 47 approvals a year. Set that trickle of new medicines against everything already approved and reordered daily: the U.S. retail pharmacy market filled close to seven billion prescriptions in 2024, inside a market valued at roughly 676 billion dollars.

Trade-offs and what's next

The clearest, longest-running tension in this system is price. Patent protection exists to let a company that spent a billion dollars or more developing a drug, most of it on candidates that never reached the market, recover that investment and fund the next round of research, the core argument for why new drugs launch at prices that can run into the tens or hundreds of thousands of dollars a year. Set against that is the plain fact that patients who need a drug today cannot wait for a patent to expire, and pharmacy benefit managers, rebates, and list prices interact in ways even trained health economists struggle to fully trace, let alone the patient wondering why the same drug costs one price with insurance and a different one without. Neither side of that argument has disappeared in sixty years of legislative attempts to referee it, and none of the PBM transparency and reform proposals currently moving through Congress fully resolve it.

Two newer forces are reshaping the discovery end of the pipeline rather than the pricing fight. AI-assisted drug discovery, using machine learning to predict how a candidate molecule will bind its target before it's ever synthesized, has moved from curiosity to active pipeline tool: Insilico Medicine's AI-designed lung fibrosis candidate reached human trials in under two years, versus roughly four by conventional methods, and Isomorphic Labs, built on the protein-structure prediction system AlphaFold, has signed drug discovery partnerships worth billions with major pharmaceutical companies. Whether that speed advantage holds through Phase 3, where most late-stage failures happen for reasons AI can't yet predict, remains unproven at scale.

Biologics and gene therapies, grown or genetically engineered rather than chemically synthesized, from antibody treatments to one-time gene therapies like Zolgensma and personalized CAR-T cancer treatments, are a newer category with manufacturing and distribution problems the small-molecule pill was never built for. A CAR-T therapy is made individually for each patient from their own cells, so it cannot be mass-produced or stockpiled; one manufacturing run serves exactly one person, and quality control has to verify each one separately. Some also need cold-chain handling as demanding as anything in Chapter 10, compounding manufacturing complexity with logistics complexity. List prices of one to two million dollars for a single dose are the direct financial consequence of that structure, pushing insurers toward new payment models, including installment plans tied to whether the therapy keeps working, because nothing about traditional drug pricing was designed for a treatment given once.

Back to the counter

The three minutes at the pharmacy counter are, in a real sense, the cheapest and fastest part of the entire chain. A pharmacist confirming your name, a technician scanning a barcode, a label matching a bottle: none of that took more than a few minutes to build in the moment, but every one of those steps sits on top of a target identified in a lab, a molecule optimized over years, volunteers who took an experimental drug so a statistician could tell whether it actually worked, a federal reviewer who read the resulting data line by line, a manufacturing plant inspected on a recurring schedule, and a regulatory framework that exists, in its current form, specifically because a reviewer named Frances Kelsey refused to be rushed in 1960 and 1961. The pill bottle doesn't carry any of that history on its label. It doesn't need to. The label just needs to be right, which is the one thing the entire system upstream of it was built to guarantee.

The leap: what it replaced, and the work behind it

For most of the nineteenth and early twentieth centuries, a bottle from the pharmacy carried no guarantee at all. Until the Harrison Narcotics Act of 1914, there was no federal law regulating morphine or cocaine, and patent medicines with secret formulas were sold openly to cure whatever the label claimed: cancer, tuberculosis, colic in infants. Mrs. Winslow's Soothing Syrup, marketed to quiet fussy babies, was morphine and alcohol. Even after the 1906 law forced ingredient labels, nobody had to prove a drug was safe to swallow. That gap killed people. In the fall of 1937, the S. E. Massengill Company dissolved a new antibiotic into diethylene glycol, a chemical cousin of antifreeze, to make a sweet raspberry liquid. Elixir Sulfanilamide killed 107 people across 15 states in a matter of weeks, most of them children, and because no law yet required safety testing, the company had broken no rule by selling it. Congress passed the 1938 Food, Drug, and Cosmetic Act the next year, the first federal law demanding a drug be shown safe before sale.

The leap since then is hard to overstate in plain numbers. In 1900, before antibiotics existed, average life expectancy in the United States was about 47 years, and the leading killers were pneumonia, tuberculosis, and infected wounds that no doctor could treat. A cut on a finger could turn into blood poisoning and death within a week. When penicillin reached ordinary patients around 1947, mortality from infections it could touch dropped by more than half almost immediately. Today life expectancy sits near 78 years, and the infectious-disease death rate has fallen from roughly 797 per 100,000 people in 1900 to a small fraction of that. Behind each of those gains stands the long, unglamorous machinery this chapter describes: chemists, trial volunteers, statisticians, and reviewers, most of whose names never appear on any bottle.

You feel the leap most in the ordinary shape of a normal day. A child with an ear infection gets an amoxicillin suspension and is playing again in two days, where a century ago the same infection could spread to the skull. A person with high blood pressure takes one small pill each morning and never has the stroke that would have arrived by 55. Strep throat, once a gateway to rheumatic fever and permanent heart damage, is a ten-day inconvenience. None of that registers as remarkable, which is the whole point. The morning it fails looks like the world before 1938: a bottle that might be sugar water, might be poison, sold by someone with no obligation to know the difference, and no agency with the power to pull it. Every boring, effective pill is a vote that the era of Elixir Sulfanilamide stays closed, kept shut by people running tests nobody sees.

Real-world examples and recent developments

The pipeline described above keeps producing new, named cases worth knowing on their own.

  • Novo Nordisk (June 2024): facing years-long shortages of its GLP-1 drugs Ozempic and Wegovy, the Danish company announced a $4.1 billion plant in Clayton, North Carolina, its fourth in the state, part of a manufacturing buildout that has run past $6 billion across sites in Denmark, France, and the U.S. since 2023. Novo Nordisk devotes $6B to expanding production as CEO indicates more to come
  • Casgevy (exagamglogene autotemcel), developed by Vertex Pharmaceuticals and CRISPR Therapeutics (approved December 8, 2023): the first FDA-approved therapy built on CRISPR-Cas9 gene editing, treating sickle cell disease by editing a patient's own stem cells to raise fetal hemoglobin levels. It is a one-time treatment for a disease that has historically required lifelong management, a concrete case of the biologics and gene therapy pathway described above already reaching patients. FDA Approves First Cell-Based Therapies for Treatment of Sickle Cell Disease
  • Comirnaty, developed by Pfizer and BioNTech (full FDA approval August 23, 2021, after emergency authorization on December 11, 2020): the mRNA COVID-19 vaccine went from a published viral sequence to authorized use in under a year, a pace normally measured in a decade or more, by running preclinical work, manufacturing scale-up, and trial phases in parallel instead of one after another, and by regulators reviewing trial data as it accumulated instead of waiting for a single completed file. FDA Approves First COVID-19 Vaccine
  • Purdue Pharma and OxyContin (Supreme Court ruling June 27, 2024): a different kind of system failure than Vioxx, rooted in marketing rather than an undetected side effect. A Purdue affiliate pleaded guilty to a federal felony in 2007 for misbranding OxyContin as less addictive than it was; in Harrington v. Purdue Pharma, the Supreme Court threw out a bankruptcy settlement that would have shielded the Sackler family from further lawsuits, sending the case back for renegotiation years after the underlying harm had already been documented. Supreme Court blocks OxyContin bankruptcy plan

Recent developments

  • FDA's Elsa AI tool (launched June 2025): a generative AI system built for internal FDA staff, used to help summarize clinical protocols, flag inspection targets, and support safety reviews. By July 2025, reporting found it had fabricated nonexistent studies in at least some outputs, a sign that the agency reviewing AI-assisted drug discovery is now also using AI internally, with the same accuracy problems that show up anywhere else generative models get deployed. FDA's artificial intelligence is supposed to revolutionize drug approvals. It's making up nonexistent studies.
  • Orforglipron (Foundayo), Eli Lilly (approved April 1, 2026): the first oral GLP-1 receptor agonist pill that can be taken at any time of day without food or water restrictions, a formulation advance over injectable semaglutide that could reshape how the obesity-drug shortages described above eventually get resolved. 5 Notable FDA Approvals From the First Half of 2026

Glossary

Active pharmaceutical ingredient (API). The chemical or biological substance in a drug that produces its therapeutic effect, manufactured separately from the finished tablet, capsule, or injectable.

Preclinical testing. Laboratory and animal studies conducted before a drug candidate is tested in humans, assessing toxicity and basic biological activity.

Investigational New Drug (IND) application. The FDA filing required before a company may begin testing a drug candidate in human clinical trials.

Clinical trial phases (1, 2, 3). The three-stage human testing sequence: Phase 1 tests safety and dosing in a small group, Phase 2 tests effectiveness and dosing in a larger patient group, and Phase 3 confirms efficacy and monitors side effects in a large, often multi-thousand-patient study.

New Drug Application (NDA). The complete regulatory submission a company files with the FDA after successful clinical trials, requesting approval to market a new drug.

Kefauver-Harris Amendment. The 1962 U.S. law, passed after the thalidomide tragedy, requiring drug manufacturers to prove both safety and effectiveness before a drug can be approved.

Current Good Manufacturing Practice (CGMP). FDA-enforced standards governing the facilities, equipment, and quality controls used to manufacture drugs, applied continuously for as long as a drug stays in production.

Hatch-Waxman Act. The 1984 law that created the modern U.S. framework for generic drug approval, including the faster Abbreviated New Drug Application pathway, in exchange for extended patent protection for brand-name drugs.

Bioequivalence. A demonstration that a generic drug delivers its active ingredient into the bloodstream at essentially the same rate and extent as the original brand-name product.

MedWatch / FAERS. The FDA's voluntary and mandatory adverse event reporting system, used to detect drug safety problems that emerge only after a drug is already on the market.

Pharmacy benefit manager (PBM). A company that negotiates drug prices and rebates on behalf of health insurers and determines which drugs a health plan covers and at what cost to patients.

Mechanism of action. The specific biochemical interaction, such as binding a receptor or inhibiting an enzyme, through which a drug produces its effect in the body.

Drug utilization review. The legally required check a pharmacist runs against a patient's medication record before dispensing, screening for interactions, duplications, and dosing errors.

Biologics and gene therapies. Drugs grown from living cells or based on genetic material rather than chemically synthesized, including antibody treatments, CAR-T cancer therapies, and one-time gene therapies, each with manufacturing and distribution demands distinct from conventional pills.

Sources and notes

Open questions

  • Estimates of the average cost to develop a new drug vary by more than a billion dollars between industry-funded studies (Tufts CSDD) and publicly-funded academic analyses (the 2020 JAMA study); this chapter presents both rather than treating either as settled.
  • How much AI-assisted drug discovery actually shortens the costliest, latest-stage trial failures, rather than just the earlier discovery phase, is still an open, actively studied question at the time of writing.
  • Whether current U.S. drug shortage policy and PBM reform proposals moving through Congress will meaningfully change the concentration of generic sterile-injectable manufacturing was still unresolved as of mid-2026.

Next, a different kind of hidden system, one built to respond in minutes instead of decades. How emergency services respond after a 911 call 👉