Artificial intelligence (AI) agents are emerging in scientific research as systems that can transform the accumulated decisions of scientists into traceable workflows. In fact, modern scientific practice depends on continual coordination of knowledge dispersed across literature, data, theory and experiment, which, however, is often made through manual, case-by-case judgement. This Primer introduces agentic AI systems as a general method to structure this process through natural language interfaces, serving as a computational brain in the laboratory. Central to this framework are practical workflows for in-depth literature search and knowledge mining, orchestration of physics-based simulations, and integration with automated experimental platforms. We begin by describing the foundations of AI agents, including task decomposition, memory, tool use and communication between agents, and by distinguishing single-agent and multi-agent designs, as well as fully autonomous and human-in-the-loop configurations. Equally important are the means by which these systems are evaluated, documented and reported reproducibly. Finally, we examine major bottlenecks, including hallucination, tool misuse, limited interpretability and safety risks, and outline a future in which coordinated networks of agents, instruments and researchers transform scientific discovery from isolated tasks into continuous adaptive experimentation. Agentic artificial intelligence (AI) systems enhance scientific workflows by structuring knowledge coordination through natural language interfaces. In this Primer, Feng et al. examine the challenges and opportunities involved in integrating AI agents in the laboratory.

Gibberellins (GAs) influence cell division and elongation, profoundly shaping plant architecture and yield. GA perception occurs when bioactive GAs bind the receptor GID1, promoting DELLA degradation and activating transcriptional programs. While GA signalling in the root endodermis is essential for promoting root elongation, the contributions of other layers have remained unclear. Here we developed a synthetic GA (sGA) that exhibits little or no binding to endogenous GID1, together with a modified GID1 (mGID1) engineered to selectively recognize sGA, enabling cell-specific activation of GA signalling in vivo. Using this system in Arabidopsis, we demonstrate that coordinated action of GA signalling in the endodermis and epidermis is required for full root elongation. Moreover, the distal spatial distribution of GA signalling and GA biosynthetic enzymes supports a model in which bioactive GAs are transported from GA3ox-expressing tissues to the endodermis and epidermis to coordinate root elongation. The sGA–mGID1 system provides a versatile platform for spatially precise reprogramming of hormone signalling, enabling synthetic control of developmental processes such as root–shoot growth balance, thereby advancing applications in plant synthetic biology and sustainable crop improvement. A synthetic gibberellin (GA)-receptor system enables spatial and temporal control of GA signalling, revealing that coordinated GA signalling in the endodermis and epidermis is required for full root elongation and providing a tool for resolving hormone action.

Shotgun metagenomes from complex environments such as soil uncover vast biodiversity. Yet most short reads produced by shotgun sequencing cannot be taxonomically or functionally annotated, as they lack a sufficiently comprehensive reference, obscuring the true structure and function of microbial communities. We introduce BBERT, a nucleotide large language model optimized for short reads. Testing on a large cohort of soil metagenomes, we found that BBERT identifies bacterial sequence syntax without relying on reference databases, enabling accurate assignment of taxonomic domain, coding potential, and reading frame directly from reads as short as 100 bp. BBERT is small and fast enough to analyze metagenomes using a modest GPU and can be used to convert short metagenomic reads directly to bacterial amino acid sequences for downstream applications. BBERT also improves de novo metagenomic assembly, reducing mismatches and gaps while accelerating runtime. Using metagenomes from wild legume nodules, we demonstrate that BBERT filtering improves bin quality while significantly accelerating de novo assembly. By providing fast, reference-free classification of short reads, BBERT unlocks large metagenomic archives for more accurate ecological and evolutionary analyses.

Characterizing noncoding variant function remains an important challenge in human genetics. Genomic deep learning models have emerged as a promising approach to enable in silico prediction of variant effects. These models include supervised sequence-to-activity models, which predict molecular phenotypes such as genome-wide chromatin states or gene expression levels directly from DNA sequence, and self-supervised genomic language models. Here we review progress in leveraging these models for noncoding variant effect prediction. We describe practical considerations for making such predictions and categorize the types of ground-truth data used to evaluate variant effect predictions, providing insight into the settings in which current models are most useful. Our Review highlights key considerations for practitioners and opportunities for improvement in model development and evaluation. This Review outlines progress in leveraging genomic deep learning models for the prediction of noncoding variant effects in humans, highlighting key considerations for practitioners, current performance trends and opportunities for improvement.

DNA represents a compelling medium for data storage due to its density, durability, and compatibility with biological systems. Current in vivo DNA data storage strategies are largely restricted to single-population systems, limiting scalability and access control. Here, we present DNA CODEC, a multi-population in vivo DNA data storage platform that combines high-density encoding with molecular obfuscation across engineered human cell populations. CODEC uses 10-ary Huffman coding with degenerate bases to optimize compressibility while maintaining decoding accuracy despite biological and technical noise. Data payloads are stably integrated into the human genome using CRISPR-based targeted insertion. Obfuscation is introduced through heterogeneous populations containing decoy clones with structurally similar, non-informative masking sequences to conceal the true payload. CODEC stably maintains genomic payloads across extended cell culture, demonstrating robustness to replication errors and population drift. Together, these features establish a scalable framework for high-fidelity and secure DNA data storage in biological systems. DNA is a dense and durable storage medium for digital information but its use in living cells remains difficult to scale and protect. Here, authors present DNA CODEC, a CRISPR-based system that distributes compressed data across engineered cell populations for stable storage and controlled access.

Deep soils store over half of terrestrial organic carbon, yet the viral mechanisms governing carbon persistence in subsurface contexts remain unclear. Here, by conducting a soil profile survey (0–700 cm) across four agroecosystems and a global synthesis of subsurface metagenomes, we established the Deep Soil Viral Resources Database, comprising 50,949 viral operational taxonomic units and 3,902 auxiliary metabolic genes (AMGs). Viral richness increased significantly with depth, accompanied by a shift towards lysogeny driven by declining carbon availability. The deep-soil virome carried AMGs and lysozymes involved in polymer degradation and cell endolysis. Viral endolysin significantly promoted necromass carbon accumulation, which contributed to increases in mineral-associated organic carbon. A synthesis of 138 global soil profiles further revealed a positive effect of viral abundance on soil organic carbon. Our findings establish a framework where viruses strengthen the microbial carbon pump in deep soils by facilitating host carbon scavenging via AMGs and promoting necromass formation via endolysin activity. Combining a metagenomic analysis of agroecosystem soils along a 7-metre depth profile with microcosm incubations and a global synthesis, the authors find increases in viral diversity, lysogeny and necromass carbon accumulation in deeper soils.

The coral microbiome is believed to play a crucial role in sustaining corals and enabling adaptation to perturbations. A ubiquitous group of coral-associated bacteria, known as Endozoicomonas, are hypothesized to provide essential metabolites to corals, although detailed dynamics of the interactions between host and microbe are poorly understood. Here multi-omics and laboratory validation experiments reveal that Endozoicomonas encodes genes for the metabolism of coral-derived testosterone and converts ubiquitous cholesterol into testosterone and progesterone, both important steroids that accumulate within the coral holobiont during thermal stress. Field sampling along a natural temperature gradient showed general suppression of coral steroid hormones under thermal stress and during white syndrome manifestation. Provision of testosterone or progesterone to corals during laboratory thermal stress induced transcriptional reprogramming of the coral innate immune response, suggesting microbial involvement in host immune modulation. These findings highlight a previously unrecognized potential mechanism of microbial contribution to coral resilience. Multi-omics and laboratory experiments reveal that a coral-associated bacterium can produce testosterone and progesterone; these two hormones alter the coral host’s innate immune responses during thermal stress.

The accumulation of persistent plastics in the ocean is causing wide-ranging environmental problems. Some commercially available compostable plastics, developed to address these issues, degrade efficiently under industrial composting conditions; however, their degradation in marine environments, particularly in the deep sea, remains limited. Here we demonstrate that thermally embedding enzymes into compostable polyesters (poly(butylene succinate) (PBS), poly(butylene succinate-co-adipate) (PBSA), poly(ε-caprolactone) (PCL) and poly(butylene adipate-co-terephthalate) (PBAT)) significantly accelerates their degradation in both coastal and deep-sea environments from water depths of 2–6 m to 5,503 m. Enzyme-embedded polyesters achieved near-complete degradation at a deep-sea depth of 757 m within 5 months. Furthermore, moderate biodegradation of enzyme-embedded polyesters was observed under extreme conditions on deep-sea floors of 5,503 m. In coastal seawater, films embedded with trace amounts (0.1 wt%) of enzyme underwent rapid degradation. The mechanical properties of enzyme-embedded polyesters were similar to those of enzyme-free polyesters. Chemical analysis confirmed the formation of low-molecular-mass degradation products and biochemical oxygen demand measurements supported the mineralization of the enzyme-embedded polyesters. The enzyme-embedding strategy enables enhanced marine degradation of compostable polyesters and represents a potential solution to marine plastic pollution. Near-complete deep-sea degradation of enzyme-embedded compostable polyesters shows that thermal enzyme embedding greatly accelerates breakdown across marine environments while preserving material performance, offering a promising strategy to mitigate persistent ocean plastic pollution.

Plant-based meat alternatives (PBMAs), designed to mimic the sensory properties of meat, may help reduce animal-sourced protein consumption and support a shift towards plant-based diets. Here we integrated PBMAs into a land-use model to assess their effects on nutrition, greenhouse gas emissions and food system sustainability. PBMA adoption not only substantially reduced agricultural land expansion and greenhouse gas emissions from the agriculture, forestry and other land use sector but also increased risks of micronutrient inadequacy. Across regions, PBMA diets exhibited diminishing marginal returns. Replacing ruminant meat alone achieved nearly half of the potential emission reductions while incurring relatively small nutritional costs. By contrast, broader substitution of livestock products delivered additional environmental benefits but markedly increased micronutrient deficiencies. Targeted policy support, sustainable land management and improved PBMA formulations will be essential to ensure equitable dietary transitions and progress towards global climate and sustainability goals. Substitution of ruminant meat with plant-based meat alternatives can support lower food system greenhouse gas emissions and limit agricultural land expansion, but micronutrient inadequacy may arise from this dietary shift.