AI Can Design Antibiotics — But Nature Still Holds the Blueprints

Last month, the San Francisco Chronicle published a sobering investigation into the antibiotic resistance crisis. The story opened with a 55-year-old woman at UCSF whose blood infection evolved resistance to multiple drugs in a matter of days, not generations, not years, but days. Doctors watched in real time as bacteria outpaced medicine.

The article's central message was hopeful: artificial intelligence may be our best weapon. Researchers at Stanford and MIT are now using generative AI to design entirely new molecules — compounds that have never existed in nature — capable of killing drug-resistant pathogens. It's a stunning technological achievement. AI can explore chemical spaces far larger than any library of existing compounds, generating millions of candidates in the time it once took to test a few dozen.

So where does that leave those of us looking for antibiotics in the soil?

The answer, I believe, is that AI and nature aren't competing solutions. They're complementary... and we need both.

What AI Does Well

AI excels at designing molecules from scratch. By learning patterns from existing drugs, machine learning algorithms can propose novel structures optimized for specific targets. Recent work from MIT's Antibiotics-AI Project has already yielded promising candidates like halicin and abaucin — compounds that work through mechanisms distinct from traditional antibiotics.

This approach has clear advantages. It's fast. It's scalable. And it doesn't depend on finding the right organism in the right place at the right time. AI can generate candidates on demand, without waiting for nature to evolve them. But there's a catch.

What AI Can't Do (Yet)

AI designs molecules based on what it has learned, which means it's fundamentally limited by the data it's trained on. Most AI models learn from existing chemical libraries, existing drugs, and existing knowledge of how antibiotics work.

What they can't easily do is invent a mechanism of action no one has ever seen.

Nature, on the other hand, has been running this experiment for billions of years. Soil bacteria have been locked in chemical warfare with each other since before multicellular life existed. The compounds they produce aren't designed to be elegant or efficient — they're designed to kill competitors in ways that are hard to evolve resistance against.

This is why 70% of our existing antibiotics trace their origins to soil microbes, particularly Actinobacteria like Streptomyces. These organisms didn't just stumble onto one good solution, they evolved entire arsenals of bioactive compounds, many of which we haven't yet discovered.

The Case for Soil

Here's what makes soil microbial discovery so powerful:

1. Novel mechanisms we haven't imagined. AI can only optimize for targets we already understand. Soil microbes may produce compounds that work through pathways we've never considered — mechanisms that AI wouldn't think to look for because they aren't in the training data.

2. Evolutionary pressure-testing. A molecule designed by AI still needs to be tested against real bacteria in real conditions. Soil-derived compounds have already survived millions of years of evolutionary pressure. They've been refined by the harshest selection process imaginable: kill or be killed.

3. The untapped reservoir. A 2025 study from Rockefeller University found hundreds of previously unknown bacterial genomes — and two novel antibiotic compounds — from a single forest soil sample. We've barely begun to explore what's out there. Most soil bacteria can't even be cultured in a lab, which means most of their chemistry remains invisible to traditional drug discovery.

4. Complexity and synergy. Soil microbes often produce not just one compound, but suites of interacting molecules. Some may work synergistically, enhancing each other's effects in ways that would be difficult to design from scratch.

Why Rewilding Matters

This is where the Nature's Pharmacy Project comes in.

If soil is a reservoir of undiscovered antibiotics, then the health of that soil determines what's available to find. Degraded soils — those affected by agriculture, development, or fragmentation — lose microbial diversity. The antibiotic-producing Actinobacteria that once thrived there decline or disappear. We're not just losing ecosystems when forests are cleared or fragmented; we're losing pharmaceutical potential we never knew existed.

Rewilding offers a way to reverse this. As forests recover from agricultural abandonment, their soil microbial communities regenerate. Research shows that bacterial diversity can recover substantially within 10–20 years of forest regrowth. The coarse woody debris, deep litter layers, and structural complexity of mature forests create conditions where antibiotic-producing microbes flourish.

The hMRI is designed to help land managers identify which recovering forests have the highest microbial potential. By measuring simple habitat features like canopy cover, litter depth, and distance from forest edges, we can estimate which sites are ready for pharmaceutical exploration and which need more time or intervention to recover.

This isn't about replacing AI. It's about ensuring that AI has the richest possible dataset to learn from — and that we don't lose irreplaceable natural compounds before we ever discover them.

A Partnership, Not a Competition

The future of antibiotic discovery isn't AI or nature. It's AI and nature.

AI can accelerate the optimization and testing of candidates. It can screen millions of possibilities and identify the most promising leads. But nature provides the raw material — the evolutionary experiments, the novel mechanisms, the chemical diversity that no algorithm could invent from scratch.

Soil microbes are the original innovators. They've been solving the antibiotic resistance problem for eons. Our job is to protect the ecosystems where they thrive, develop tools to identify the richest sites, and ensure that when AI goes looking for inspiration, nature still has something to teach.

The woman in the Chronicle story survived because doctors had a fourth-line antibiotic to try. Someday, that drug might fail too. When it does, the next solution might come from an AI model trained on millions of compounds — or it might come from a handful of forest soil, collected from a rewilded hillside that someone decided was worth protecting.

We need both.

The Nature's Pharmacy Project is developing the Habitat-Based Microbial Recovery Index (hMRI) to help conservation practitioners assess soil microbial recovery using simple field measurements. Learn more at naturespharmacyproject.com.