Latent Labs has launched Latent-Y, a new autonomous artificial intelligence agent designed to carry out therapeutic antibody design campaigns from a text prompt, research plan, or scientific paper, marking an ambitious step deeper into automated drug discovery. The London- and San Francisco-based company said the system can move from a research goal to lab-ready antibody sequences in hours, compressing work that would otherwise take specialists weeks. The release positions Latent-Y drug design as an expert-level workflow layer built on Latent-X2, Latent Labs’ earlier model for designing antibodies and peptides with drug-like properties. Public material from Latent Labs confirms that Latent-X2 was introduced in December 2025 as a frontier model for generating high-affinity antibodies and macrocyclic peptides with developability and low ex vivo immunogenicity in mind.
According to the company’s announcement, Latent-Y is intended to work much like a computational protein design expert operating inside a digital research environment. Latent Labs said the agent can analyze target molecules, identify potential epitopes, design antibody candidates using Latent-X2, validate them computationally, and iterate until design goals are met. The company also said scientists can choose whether to let the system run end to end or stop at each stage for review, with every design decision logged for inspection. That framing makes autonomous antibody design the central promise of the launch: not simply a model that proposes molecules, but an agent that runs an entire design campaign. The product is currently being opened to selected partners rather than to the broad public.
How Latent-Y builds on Latent-X2
The scientific foundation for Latent-Y rests on Latent-X2, which Latent Labs previously described as an all-atom generative model conditioned on target structure, epitope specification, and optional antibody framework. In public materials and its technical report, the company said Latent-X2 was able to generate binders against 9 of 18 evaluated targets using only 4 to 24 designs per target, with reported affinities ranging from picomolar to nanomolar. The report also said the model aimed to produce designs that were not only potent but also more developable and less immunogenic than many earlier AI-generated candidates. Those claims matter because Latent-Y drug design depends on Latent-X2 not merely as a brainstorming engine, but as the molecule-generation core that converts reasoning into candidate therapeutics.
The new launch extends that platform by adding an orchestration and reasoning layer. In the company’s description, Latent-Y can take a high-level therapeutic goal in natural language, use outside tools and biological databases, and autonomously decide how to progress through target analysis, epitope selection, sequence design, and computational checks. One notable claim in the announcement is that, during a cross-species binder campaign, the system reportedly extended its own capabilities by implementing a custom generative method from a brief natural-language instruction to solve a challenge it had not been explicitly built for. That is a striking assertion, though it should still be read as a company-reported capability until independent benchmarking and peer-reviewed validation are more broadly available.
Lab-validated results are the strongest part of the announcement
The most important part of the launch is the claim that Latent-Y has already shown lab-validated results across three antibody design campaigns without human filtering or intervention. In the material you provided, Latent Labs said those campaigns included epitope discovery, cross-species binder design, and antibody design from a published paper targeting the human transferrin receptor for blood-brain barrier crossing. The company further said the campaigns produced a 67% target-level success rate and single-digit nanomolar affinities. Those results, if they continue to hold up in external testing, would place the system in a serious position within the growing AI drug discovery field, where many platforms can generate molecules computationally but fewer can point to wet-lab confirmation across varied workflows. The caution is that these findings currently appear to be coming from company disclosures and trade coverage, rather than an independent, peer-reviewed Latent-Y paper available in the public domain.
That distinction matters because biotech announcements often combine genuine technical progress with forward-looking framing. For example, Latent-X2 has a technical report and preprint-level detail in the public domain, which gives outside readers something concrete to examine. By contrast, Latent-Y’s launch appears to be newer and less exhaustively documented in public primary literature so far. That does not invalidate the claims, but it does mean the strongest evidence currently available is still company-led. For readers tracking Latent Labs launch announcements, the balanced conclusion is that the product looks credible in light of Latent-X2’s documented foundation, while the larger claims about full autonomy and generalizable protein design will become easier to judge as more external validation emerges.
Why Latent-Y matters in the AI drug discovery race
Latent-Y enters a crowded and fast-moving field in which biotechnology companies are trying to show that artificial intelligence can do more than accelerate prediction. The real commercial and scientific prize is to prove that AI can design therapeutically useful molecules with enough speed, quality, and reproducibility to change how discovery teams operate. Latent Labs is explicitly making that case. The company says one researcher can run multiple design campaigns simultaneously across targets and modalities, and that expert users in its studies completed campaigns 56 times faster with the agent than independent expert estimates would suggest. That makes AI drug discovery and protein design agent searches especially relevant for this story, because the launch is really about workflow transformation as much as molecule generation.
Simon Kohl, founder and chief executive of Latent Labs, said in the company announcement that Latent-X2 provided the breakthrough of computationally designed antibodies with drug-like developability, while Latent-Y adds an expert reasoning layer to handle the full workflow autonomously. Recast in indirect speech, Kohl argued that this combination could let a single researcher run dozens of campaigns in parallel and act as a true force multiplier for discovery teams. That message is strategically important because it shifts the value proposition from isolated model performance to team productivity at scale. In other words, the company is not just claiming better antibody design. It is claiming a new operating model for therapeutic research.