I would never pretend to know everything about the inner workings of a monoclonal antibody. However, I construct things, whether be it buildings or biological structures and they both have taught me the same lesson which is the biggest leaps do not come from working harder within an existing system. They come from questioning the system itself.

That is what AI is doing to drug discovery right now. And as someone working at the intersection of biology and industry through TerraPHA, it is the development I find most worth paying attention to.

Developing a single drug, from target identification to regulatory approval, used to take 12 to 15 years on average. 

And the cost? 

It was somewhere between $1.5 and $2.6 billion, depending on the therapeutic area. And the success rate from preclinical stage to approval sat below 10%. 

The inefficiency was not born from laziness. It was structural. Biology is extraordinarily complex. The number of potential molecular combinations that might engage with the disease target is countless, much more than what any group of scientists can reasonably explore in their lifetimes. Traditional drug discovery was, in a sense, always underpowered for the problem it was trying to solve.

AI does not just make the old process faster. It changes what the process is capable of.

The Most Transformative Application Is in Molecular Design.

AI models, particularly generative models trained on massive datasets of known compounds, protein structures, and biological interactions can now propose novel molecules that fit a target with a specificity and speed that was unimaginable five years ago. DeepMind’s AlphaFold cracking the protein structure prediction problem in 2020 was the inflection point most people point to, and rightly so. It opened a door that the entire industry has been walking through ever since.

But protein folding is just one chapter. AI is now being used to predict drug toxicity before a single compound is synthesised, to identify patient subpopulations most likely to respond to a therapy, and to repurpose existing approved drugs for new indications, reducing years of exploratory work into weeks.

The global AI-in-drug-discovery market is on track to cross $11 billion by 2028, growing at close to 40% annually. That is not just hype, that is pharmaceutical companies, biotech startups, and research institutions restructuring their entire discovery pipelines around these tools because the output is real.

Why This Matters Beyond Pharmaceuticals

Here is where my own work at TerraPHA becomes relevant and where I think the healthcare conversation is still too narrow.

We use bio-based systems to solve problems in aquaculture, animal nutrition, and soil health. On the surface, that sits at a comfortable distance from drug development. But the underlying science is converging fast. The same machine-learning tools being used to predict protein-drug binding are being applied to understand microbial metabolic pathways, optimise fermentation conditions, and design more effective bio-inputs for agriculture and animal health.

When we work on improving gut microbiome function in fish and cattle reducing pathogen load, improving immunity without antibiotics, we are operating in the same biological domain that AI is now illuminating so rapidly. The distance between what we do and what a drug discovery team does is reducing at the molecular level, even if the applications look different on the surface.

Several of the tools being used to develop novel therapeutics are the same tools being explored for next-generation probiotic and enzyme design in industrial biotech. 

The Remaining Challenge

None of this means the hard problems are solved. AI is extraordinary at pattern recognition within the space it has been trained on. Where it still struggles is with genuine biological novelty, disease mechanisms that do not resemble anything in the training data, rare conditions with thin literature, or the complex systemic effects of a drug in a living human body across months of use.

Clinical trials, for all the inefficiency they represent, exist for a reason. Biology surprises us. The AI-assisted pipeline has compressed target identification and lead optimisation dramatically as the chart shows, those early phases are where the time savings are sharpest. But the clinical phases, where human variability takes over, remain stubbornly long. The honest projection is not that AI will eliminate drug development timelines, but that it will push the frontier of what is even worth taking to trial and that matters enormously.

A world where fewer failed trials consume resources that could go toward viable therapies is a meaningfully better world for patients. That is what is actually being built here.

Running a biotech company in India in 2026 means operating in a system that is still catching up in regulation, in infrastructure, in the deep capital markets that fund long-horizon science. 

What AI in drug discovery tells me is what it reinforces every time I read about a new compound designed in silico and validated in weeks is that the biology has always held the answers. We just lacked the tools to ask it the right questions, at the right scale, fast enough.

At TerraPHA, we are asking our own version of those questions. In ponds. In soil. In animal guts. The scale is different. The need is the same.