Agents have tools. Let’s transform human health.
This week, NVIDIA announced the BioNeMo Agent Toolkit, a collection of domain-specific tools that enables AI agents to perform scientific workflows spanning biology, chemistry, genomics, and drug discovery. We think the announcement represents something much bigger than another AI product release. It marks another milestone in the maturation of agentic AI.
In our view, every modern AI agent has three components: a model, a harness, and a set of tools.
Over the past two years, we’ve watched each of these mature in rapid succession.
The first component was the model. Frontier AI labs have delivered remarkable progress in reasoning, coding, planning, and long-context understanding. Today, increasingly capable models seem to arrive every few weeks.
The second component was the harness. Harnesses connect models with their environment. Rather than simply generating text, systems like Claude Code, OpenClaw, and more recently NVIDIA’s NemoClaw enable models to maintain context, execute multi-step workflows, recover from failures, and interact continuously with external systems. Claude Code entered preview in February 2025, OpenClaw emerged in late 2025, and NVIDIA introduced NemoClaw at GTC in March 2026. The pace of innovation has been extraordinary.
The third component is now maturing: tools.
Coding was the natural first application because the tools were already there. Compilers, terminals, version control systems, and test suites gave agents a rich environment to interact with. Once harnesses connected models to those tools, software engineering became the first domain where autonomous agents could create real value.
Biology has always needed the same transition. Scientific agents require access to protein structure prediction, molecular docking, genomic analysis, literature retrieval, experimental design, and laboratory automation. NVIDIA’s BioNeMo announcement is significant because it helps establish that scientific tool layer. Rather than reasoning abstractly about biology, agents can now interact with the computational tools biologists already use.
Perhaps the clearest sign that this ecosystem has matured is adoption. NVIDIA announced that more than 50 organizations are already using BioNeMo, including Lilly, Schrödinger, Dassault Systèmes, Databricks, Snowflake, the University of Washington Institute for Protein Design, and numerous AI-native biotechnology companies. That’s an important signal. The conversation is rapidly shifting from building agents to solving problems.
We believe this changes the bottleneck.
Once capable models, robust harnesses, and domain-specific tools exist, the limiting factor is no longer the AI stack itself.
The limiting factor becomes the problems we choose to solve.
Few problems are more important than human health.
Drug discovery routinely takes 10–15 years, costs billions of dollars per approved medicine, and roughly 90% of candidates entering clinical development fail. The largest cause of failure is insufficient efficacy in humans, followed by toxicity and poor drug-like properties. Biology is difficult because it is combinatorially complex. Humans have roughly 20,000 protein-coding genes whose products interact across diverse cell types, tissues, signaling networks, and environments. The search space is effectively impossible to explore exhaustively.
Fortunately, evolution already has.
Evolution has been running experiments for billions of years across millions of species. Every surviving species represents a successful biological solution to a unique environmental challenge. Elephants evolved remarkable cancer resistance through expanded TP53 biology. Naked mole rats evolved extraordinary cancer resistance through high-molecular-weight hyaluronan. Bowhead whales evolved mechanisms supporting exceptional longevity and enhanced DNA repair. Bats tolerate viral infections that are devastating in other mammals while maintaining surprisingly low cancer rates.
These aren’t biological curiosities.
They’re validated solutions to some of medicine’s hardest problems.
At Intertwined Biosciences, we believe one of the greatest opportunities for AI isn’t inventing biology from scratch. It’s helping us understand the biology that evolution has already discovered.
Our goal is to build systems that connect evolutionary biology, scientific literature, computational models, and experimental validation into a continuous discovery engine. We believe the next generation of medicines won’t come from searching biology at random. They’ll come from understanding the solutions that evolution has already discovered and translating them into therapies that improve human health.
Agentic AI is rapidly becoming the infrastructure that makes that future possible.