Every medicine you take could one day be designed by AI. At least, that's what Isomorphic Labs President Max Jaderberg believes.
At Bloomberg Tech in London on November 2, Jaderberg predicted that AI would eventually design all medicines.
It's an ambitious claim, especially considering where his own company stands today.
Isomorphic Labs is still developing its first drug candidates, and none has entered human clinical trials. Jaderberg described the company as being in its "first innings," suggesting there's still plenty of work ahead.
Earlier this year, CEO Demis Hassabis said Isomorphic expected to begin human trials by the end of 2026, after missing an earlier 2025 target. As of November 3, the company hasn't publicly confirmed that any trials have started.
So, how close are we to letting AI design the medicines we might someday take?
How does AI design a medicine?
Most medicines work by interacting with specific molecules in the body, often proteins involved in a disease.
The challenge is finding a chemical compound that interacts with the intended protein in the right way.
Scientists have traditionally spent considerable time designing possible compounds, testing them in laboratories and refining the most promising candidates.
Isomorphic Labs wants AI to help researchers narrow that search.
The company grew out of Google DeepMind, which developed AlphaFold, an AI system that predicts the three-dimensional shapes of proteins.
Knowing a protein's shape helps scientists understand where a potential medicine might attach. But researchers also need to know how a drug molecule will interact with that protein.
Isomorphic's drug design engine, IsoDDE, is designed to answer those questions.
Say researchers want to block a protein associated with cancer. They could use IsoDDE to identify possible attachment points and predict which molecules might bind to the protein.
The system can also propose chemical designs based on properties scientists want, such as how easily a compound dissolves.
Researchers can then choose which candidates deserve laboratory testing.
In theory, AI could help scientists spend less time investigating molecules that are unlikely to work. The next question is whether Isomorphic's technology is already delivering those results.
What has Isomorphic Labs actually achieved?
In February, Isomorphic reported that IsoDDE more than doubled AlphaFold 3's accuracy on a particularly difficult protein-drug prediction benchmark.
The test examined molecular combinations the system hadn't previously encountered, assessing whether an AI model can make predictions beyond familiar examples.
Isomorphic also reported improvements in predicting how tightly molecules attach to biological targets.
Those results suggest the system can make better predictions in certain computational tests. They don't establish how well an AI-designed drug will perform inside the human body.
Then, in September, the company described an automated drug design experiment.
Researchers gave the AI a biological target and specified the properties they wanted in a molecule. Over two to four days, the system repeatedly generated potential designs and selected candidates for further evaluation.
Scientists then made selected compounds and tested them in the laboratory.
According to Isomorphic, one candidate showed activity against its intended biological target.
That's encouraging for the company's drug discovery approach. However, a compound that performs well in an experiment may behave very differently in the human body.
Could AI make drug discovery cheaper and faster?
Here's where the technology could have a considerable impact.
Drug discovery often involves testing numerous chemical compounds before researchers find one worth developing further. Many candidates are rejected because they lack the desired properties or produce unwanted effects.
AI could help scientists evaluate more possibilities on a computer before spending money on physical experiments.
For example, a pharmaceutical company investigating a difficult-to-treat disease could use AI to search for molecules that traditional discovery methods might overlook.
Researchers would still need to manufacture and test promising compounds, but a more targeted search could reduce some of the experimental work required early in development.
As The Neuron previously reported on Microsoft's Quine, AI systems are starting to connect computational predictions with laboratory experiments. Of course, saving time during early discovery doesn't guarantee that a medicine will reach patients sooner.
Drug development involves several stages beyond designing a molecule, and setbacks can happen long after researchers identify a promising candidate.
Big pharmaceutical companies are already interested
Isomorphic isn't pursuing its vision entirely on its own.
Through research partnerships, the company is working with Novartis and Eli Lilly to discover potential medicines against selected biological targets.
Novartis expanded its collaboration in 2025. Isomorphic also announced an agreement with Johnson & Johnson in January 2026 to investigate small-molecule drugs and larger biological treatments.
These collaborations give Isomorphic access to pharmaceutical research expertise while allowing drugmakers to explore its AI technology.
However, the partnerships haven't been reported to produce approved medicines.
The companies still face the same scientific question: can the compounds that AI designs become safe and effective treatments?
Designing a drug is only the beginning
Before a medicine reaches patients, researchers need evidence that it works and that its risks are acceptable.
Under the FDA's drug development process, promising candidates undergo preclinical testing, which includes laboratory and animal studies.
If the results support further investigation, researchers can seek authorization to begin human clinical trials.
Those trials examine safety and effectiveness, while regulators review the evidence before deciding whether a treatment can be marketed.
Even an AI system that predicts molecular interactions accurately can't establish whether a drug will cause unexpected side effects or improve a patient's health.
A promising compound might also break down too quickly in the body, preventing it from reaching the intended target.
As AI takes on more of the design process, scientists could devote more attention to checking predictions and investigating how compounds behave in living systems.
For Isomorphic Labs, the next milestone will be getting its first candidates into human trials. After that, the company will need clinical evidence to demonstrate whether its approach produces treatments that work.
Jaderberg believes AI will eventually design every medicine, but the technology's success will depend on something patients can actually benefit from.
The real breakthrough will come when AI-designed drugs prove safe and effective in humans, and researchers can show that the technology helps those treatments reach patients sooner.
