Meaning in the Age of AI

The Race Inside Your Cells

AI is rewriting drug discovery. Here's when it reaches the diseases that matter most to you.

Scroll to explore ↓
Norman Rockwell mixed-media. A young researcher, East African with deep brown skin, oil-painted in warm detail, peering...

A new drug takes, on average, 12 years to travel from a laboratory hypothesis to your medicine cabinet. AI is compressing that timeline. The question for anyone paying attention is how fast, and for which diseases first.

In 2023, a pharmaceutical startup called Insilico MedicineHong Kong-based biotech company that uses generative AI for drug discovery. Their candidate ISM001-055 reached human trials in under 18 months from target identification. announced something that would have sounded absurd five years earlier. They had identified a novel drug target for a fibrotic disease, designed a molecule to hit it, and moved that molecule into human clinical trials in under 18 months. The traditional version of that process takes four to six years, and most candidates die somewhere along the way.

That single result is an anecdote. What followed was a pattern. By 2025, the number of IND filingsInvestigational New Drug applications, submitted to the FDA before a new drug can be tested in humans. The filing marks the transition from lab research to clinical trials. for AI-originated molecules hit its highest single-year jump on record. Hundreds of AI-designed drug candidates entered clinical pipelines across oncology, fibrosis, autoimmune disorders, and rare diseases. The pharmaceutical industry had been talking about AI for years. In 2025, it started betting real money.

The Speed Gap

Drug discovery has always been a problem of scale. AI changes the math.

The human body contains roughly 20,000 protein-coding genes. Each protein folds into a specific three-dimensional shape, and that shape determines how it behaves, what it binds to, and what goes wrong when it misfolds. For decades, determining a single protein's structure could take a PhD student's entire graduate career. AlphaFoldDeepMind's protein structure prediction system. AlphaFold 2 solved the protein-folding problem in 2020. AlphaFold 3, released in 2024, predicts interactions between proteins, DNA, RNA, and small molecules. predicted the structure of nearly every known protein in the human body, and its third version extended that prediction to how proteins interact with potential drug molecules, showing a 50% improvement over previous computational methods for protein-ligand interactionsThe binding between a protein and a small molecule (ligand). Understanding these interactions is central to designing drugs that attach to the right target without triggering side effects..

In parallel, generative AI models trained on billions of known molecular structures learned to propose new ones. Feed the model a target protein shape and a set of desired properties, and it generates candidate molecules that might bind effectively. A process that required teams of medicinal chemists screening thousands of compounds became, in some cases, a computation that ran overnight.

12 yrs
Traditional drug development timeline, target to approval
~4 yrs
AI-accelerated timeline projected by leading biotech firms
$2.6B
Average cost to develop one approved drug traditionally

The speed matters for an obvious reason: every year a drug spends in development is a year patients go without it. But the cost matters almost as much. When each approved drug carries a $2.6 billion price tag, pharmaceutical companies focus on diseases with the largest markets. Rare diseases, affecting fewer than 200,000 people in the U.S., get less investment because the return is harder to justify. AI's ability to cut discovery costs could change that calculation entirely.

What's Already Here

AI has already reshaped early-stage drug discovery. The evidence is in the pipeline numbers.

AI-Discovered Drug Candidates in Clinical Trials
By therapeutic area, as of early 2026
Oncology
~150 candidates
Immunology
~80
Neurology
~65
Fibrosis
~45
Rare Diseases
~35
Infectious Disease
~25

Oncology dominates for the same reason it dominates traditional pharma: cancer is many diseases, each requiring its own approach, and the market is enormous. But the more revealing signal is the growth in rare diseases and fibrosis, areas where traditional drug economics make investment difficult. AI lowers the cost of exploring those targets enough that smaller biotech companies can take the risk.

In 2020, researchers at MITMassachusetts Institute of Technology. Their lab used AI to screen 107 million chemical compounds in days, identifying halicin, a structurally novel antibiotic effective against drug-resistant bacteria. used a machine learning model to screen 107 million chemical compounds and identified halicinAn antibiotic compound discovered by AI in 2020. Named after HAL from 2001: A Space Odyssey. Effective against several drug-resistant bacterial strains in lab tests., an antibiotic with a mechanism unlike any existing drug. The screening took days. A human team testing compounds one by one would have needed decades. That paper was a proof of concept. Five years later, AI-driven antibiotic discovery has produced multiple candidates moving toward clinical testing.

Norman Rockwell mixed-media. An elderly man sitting in a clinic waiting room, oil-painted with tender realism. He holds...

The Map, Disease by Disease

Every disease sits at a different point on this trajectory. Some are years away from AI-assisted breakthroughs. Others are months.

2025–2027
Drug-Resistant Infections
AI-discovered antibiotics already in late preclinical and early clinical stages. The pipeline is the furthest along because the problem is well-defined: find molecules that kill specific bacteria without harming human cells. AI excels at this kind of targeted search.
2026–2029
Certain Cancers (Targeted Therapies)
AI-designed small molecules targeting specific oncogenic mutations are entering Phase II trials. Cancers with clear genetic drivers, like certain lung and breast cancer subtypes, will see AI-originated treatments reach patients first.
2027–2031
Fibrotic Diseases
Idiopathic pulmonary fibrosis and liver fibrosis have few effective treatments. Insilico Medicine's IPF candidate is the canary in this coal mine. If it succeeds in Phase II, expect a wave of AI-designed fibrosis drugs behind it.
2028–2033
Rare Genetic Diseases
AI is making it economically viable to develop drugs for diseases affecting tiny patient populations. Orphan drug development times could shrink from 15+ years to under 7, making dozens of previously "undruggable" conditions targetable.
2029–2035
Neurodegenerative Diseases
Alzheimer's, Parkinson's, and ALS remain the hardest targets. The biology is complex, and even finding the right target to aim at is uncertain. AI helps narrow the search, but the clinical trials are long and the failure rate is steep. As of early 2025, 182 clinical trials were assessing 138 drugs in the Alzheimer's pipeline alone.
2030–2040
Autoimmune Disorders
Conditions like lupus and rheumatoid arthritis involve the immune system attacking the body's own tissues. The challenge is modulating immunity precisely enough to stop the damage without leaving the patient vulnerable to infection. AI helps model these interactions, but clinical validation is slow.

These timelines represent when AI-originated or AI-substantially-assisted treatments could reach approved status for patients. The range reflects genuine uncertainty: clinical trials are the bottleneck that AI can compress but cannot eliminate. A drug still has to prove it works in human bodies, and human bodies still take time to respond.

Where the Leverage Is

AI contributes differently at each stage of drug development. Its impact is largest where the search space is widest.

1
Target Identification
Finding which protein or gene to aim at. AI scans genomic, proteomic, and clinical data to surface targets that human researchers might not prioritize. High leverage: opens doors to previously invisible opportunities.
2
Molecule Design
Generating candidate compounds. Generative models produce novel molecular structures optimized for binding, toxicity, and synthesizability. Highest leverage: replaces years of manual screening with hours of computation.
3
Preclinical Prediction
Predicting how candidates will behave in the body before animal testing. AI models simulate absorption, metabolism, and off-target effects. Moderate leverage: reduces the number of dead-end candidates entering expensive testing.
4
Clinical Trial Design
Selecting patients, optimizing dosing, identifying biomarkers for response. AI makes trials faster by enrolling the right people and measuring the right outcomes. Growing leverage: still underutilized relative to its potential.

The pattern is clear. AI's greatest current impact is at the front of the pipeline, where the problem is one of search and pattern recognition across vast chemical and biological spaces. Its impact diminishes further along, where the problem shifts to biological uncertainty and regulatory process. Over time, that second category will compress too, but it will take longer because it requires both better models and better data from more trials.

The Hard Truth About Timelines

Faster discovery does not mean faster cures. The gap between finding a promising molecule and putting it in a patient's hand is where most drugs die.

Drug Development Attrition Rates
Percentage of drug candidates that survive each phase
Enter Phase I
100%
Enter Phase II
63%
Enter Phase III
31%
FDA Approval
~12%

Roughly 88% of drugs that enter clinical trials fail. AI improves the odds at the front end by generating better candidates and filtering out likely failures earlier. Some researchers estimate AI could push the overall approval rate from 12% to 20% or higher. That sounds incremental until you do the math: it means nearly twice as many approved drugs from the same number of clinical trials.

The bottleneck was never imagination. Chemists could always dream up molecules. The bottleneck was testing them. AI shrinks the gap between "possible molecule" and "molecule worth testing in a human," and that changes the economics of every disease.

But regulatory timelines set a floor. Phase III trials for oncology drugs average four to six years. For Alzheimer's, they can run seven to ten. AI can enrich the patient populations entering those trials and improve the statistical designs that measure outcomes, but it cannot make neurons regenerate faster or tumors respond sooner. Biology has its own clock.

Composite portrait, fictional person, real circumstances
Portrait of a person
Maria Gutierrez
62, former clinical research coordinator, San Antonio
One Person's Story

I spent twenty-three years coordinating clinical trials for a pulmonary research center. We ran studies for COPD, asthma, IPF. You learn the rhythm: recruit patients, run the protocol, watch most drugs wash out at Phase II. For every drug that made it to market, I'd seen maybe forty fail. That ratio becomes part of how you see the world. You stop getting excited early.

Last year our site was recruited for a trial with a fibrosis compound that had been designed, start to finish, by an AI platform. The timeline from target to IND was something like fourteen months. I've never seen a sponsor move that fast. The molecule looked clean in preclinical. The binding profile made sense. Our principal investigator kept saying "where's the catch" because we'd all been conditioned to expect one.

It's early. I know better than to call anything a success before Phase III data. But the thing that changed for me is the pace. If this generation of tools can double the number of candidates that make it through the gauntlet, that means I might see drugs approved for diseases I'd written off as too hard. My sister has lupus. I'd stopped believing there'd be a new treatment in her lifetime. I don't feel that way anymore.

What This Means, Personally

The impact of AI on medicine arrives unevenly. Your proximity to it depends on what disease you're watching.

If you or someone you love is dealing with a cancer that has a known genetic driver, the AI-accelerated pipeline is already producing candidates in clinical trials you might be eligible for within the next two to three years. If the condition is a rare genetic disease, AI is for the first time making it economically rational for companies to pursue treatments, but the clinical validation process still takes years once a candidate is found.

For neurodegenerative diseases, the honest answer is longer. AI is speeding up the search for viable drug targets in Alzheimer's and Parkinson's, and the scale of that search is unprecedented. But the underlying biology remains poorly understood compared to, say, bacterial infections or even cancer. The 182 active clinical trials in the Alzheimer's pipeline reflect how many different hypotheses researchers are testing simultaneously. AI helps run more of those experiments, faster. It does not resolve which hypothesis is right.

Near-Term Impact (2025–2030)
Drug-resistant infections, targeted cancer therapies, fibrotic diseases, certain rare genetic conditions. AI-originated treatments will reach patients for these categories first because the biology is better understood and the clinical endpoints are clearer.
Longer Horizon (2030–2040)
Neurodegenerative diseases, complex autoimmune disorders, multi-factor chronic conditions. AI accelerates the search but cannot shortcut the clinical validation for diseases where even defining "success" is contested.

The gap between those two columns is where the frustration lives. If your family is affected by something in the right column, the phrase "AI is transforming medicine" can feel hollow. The transformation is real. It is also slow enough, for some diseases, that it tests patience. The best framing is probably this: AI is collapsing the discovery phase of drug development from a decade to a year or two. The testing phase, where biology and regulation set the pace, is compressing more slowly. The total effect is significant. The timeline to a specific cure depends on which disease you're asking about.

The Race That Matters

Every disease is a lock. AI is generating keys faster than any tool we've ever built. Some locks will open in the next few years. Others will take a decade or more. The trajectory bends toward more treatments, for more conditions, reaching more people, sooner. That sentence would have been aspirational five years ago. The pipeline numbers make it a forecast.

Jesse Walker
Jesse Walker
Jesse Walker is a philosopher, a meditation teacher, a business founder and a father. He is optimistic about humanity’s ability to shape AI into a force for global good.