Latent Labs CEO Simon Kohl: Generative AI Enters the 'Programmable Biology' Era

Generative AI is fundamentally reshaping pharmaceutical development. Simon Kohl, founder of Latent Labs and a core researcher behind AlphaFold, argues that biology has entered a programmable era.

Speaking at CVPR 2026 in Denver, Kohl addressed the critical inefficiencies in modern drug discovery. He highlighted how current methods fail to deliver viable treatments efficiently.

The industry faces staggering costs and timelines for new medications. Most candidates never reach patients due to early-stage failures.

Kohl proposes a technical evolution moving beyond simple prediction. His vision involves autonomous agents designing molecules from scratch.

Key Takeaways from the CVPR 2026 Address

• High Failure Rates: Approximately 90% of candidate drugs fail during development phases.
• Prohibitive Costs: Bringing one new drug to market costs over $2 billion USD on average.
• Time Intensive: The process takes more than 10 years from initial research to approval.
• Root Cause: Current failures often stem from starting with incorrect molecular structures.
• New Paradigm: Shift from static structure prediction to dynamic, conditional generation.
• Autonomous Agents: Future systems will use AI agents to iterate designs without human intervention.

The Crisis in Traditional Drug Discovery

Modern pharmaceutical R&D operates on a broken model. Developers spend decades searching for effective compounds. This approach relies heavily on trial and error rather than precise engineering.

The financial burden is unsustainable for many biotech firms. Investors demand faster returns, yet the science moves slowly. This disconnect stifles innovation in treating rare diseases.

Kohl identifies a fundamental flaw in this workflow. Researchers often begin with known biological targets. They then search for molecules that might fit these targets.

This reverse-engineering process is inherently limited. It restricts the chemical space explored by scientists. Many potentially effective molecules are overlooked entirely.

Starting from the Wrong Molecules

The core issue lies in our starting point. We assume we understand the target well enough. However, biological systems are complex and dynamic.

Static models cannot capture this complexity effectively. They provide a snapshot, not a movie. This leads to designs that work in silico but fail in vivo.

Kohl’s team at Latent Labs focuses on this gap. They aim to generate molecules that are optimized for function. This requires a different computational approach.

From AlphaFold to Generative Design

AlphaFold revolutionized structural biology by predicting protein shapes. This was a massive leap forward for the field. However, prediction alone does not create new therapies.

Kohl served as a key member of the DeepMind team. He witnessed firsthand the power of accurate structure prediction. Yet, he saw its limitations in drug creation.

The next step is conditional generation. This technique allows AI to design molecules based on specific criteria. These criteria include binding affinity and safety profiles.

The Technical Leap Forward

Generative models can explore vast chemical spaces quickly. They propose novel structures that humans might miss. This accelerates the initial screening process significantly.

Unlike previous versions of AI tools, these models learn from diverse data sets. They integrate information from genomics, proteomics, and clinical trials. This holistic view improves design accuracy.

Latent Labs is building infrastructure to support this shift. Their platforms enable researchers to define desired outcomes. The AI then generates potential solutions automatically.

Latent Labs and the Latent-X1 Model

Latent Labs has developed two generations of foundational models. The first iteration laid the groundwork for generative biology. It demonstrated the feasibility of AI-driven design.

The latest release, Latent-X1, represents a significant upgrade. This model is designed for general-purpose protein generation. It handles complex constraints with high precision.

The system uses advanced diffusion techniques. These methods allow for fine-grained control over molecular properties. Researchers can tweak parameters to optimize results.

Autonomous Agents in Molecular Design

Beyond static models, Kohl envisions autonomous agents. These AI entities can plan and execute experiments virtually. They learn from each iteration to improve future designs.

This creates a feedback loop of continuous improvement. The agent refines its understanding of biological interactions. It reduces the need for manual oversight.

Such autonomy is crucial for scaling drug discovery. Human experts are scarce and expensive. AI agents can operate around the clock without fatigue.

They handle routine tasks, freeing scientists for creative work. This division of labor maximizes productivity in labs worldwide.

Industry Context and Future Implications

The shift toward programmable biology aligns with broader AI trends. Companies like Insilico Medicine and Recursion Pharmaceuticals are already leveraging similar technologies. Competition in this sector is intensifying rapidly.

Western markets are leading this transformation. Regulatory bodies in the US and Europe are adapting guidelines. They recognize the potential of AI to speed up approvals.

However, challenges remain regarding validation. Clinical trials still require rigorous testing of AI-generated compounds. Trust in algorithmic outputs must be earned through consistent results.

What This Means for Stakeholders

Pharmaceutical companies must adapt their R&D strategies. Ignoring generative AI risks falling behind competitors. Integration of these tools is no longer optional.

Biotech startups have a unique opportunity. They can leverage cloud-based AI platforms to reduce overhead. This lowers barriers to entry for innovative therapies.

Investors should look for teams with strong AI expertise. The intersection of biology and computer science is where value will be created. Partnerships between tech firms and pharma giants will likely increase.

Looking Ahead: The Road to 2030

The next five years will be critical for adoption. Early successes will drive wider acceptance of generative design. Failures may lead to stricter regulations or skepticism.

Kohl predicts a hybrid workflow will emerge. Scientists will guide AI agents using domain expertise. This collaboration ensures that generated molecules are biologically plausible.

Advancements in compute power will further accelerate progress. Specialized hardware for biological simulations is under development. This will make real-time design feasible for larger projects.

Gogo's Take

• 🔥 Why This Matters: This technology could slash drug development costs by 50% or more. It enables rapid responses to emerging pandemics by designing vaccines in weeks rather than years. Patients gain access to life-saving treatments much faster.
• ⚠️ Limitations & Risks: AI models may hallucinate non-existent biological interactions. Over-reliance on algorithms could overlook rare side effects. Ethical concerns arise regarding patenting AI-generated life forms. Data privacy remains a significant hurdle.
• 💡 Actionable Advice: Biotech leaders should pilot Latent-X1 or similar tools on small projects. Test integration with existing lab workflows immediately. Invest in training staff to interpret AI outputs critically. Do not replace human judgment; augment it.