Disclaimer: never financial or medical advice
When we think about groundbreaking scientific discoveries, we usually picture human geniuses in lab coats.
But a new breed of researchers is emerging – not human prodigies, but AI “agentic scientists” working tirelessly in labs and on computers.
These autonomous or semi-autonomous AI agents operate with surprising independence, generating hypotheses, running experiments, analyzing data, and even helping to publish findings. In short, they’re beginning to take on roles once reserved for human scientists. The rise of these agentic AI researchers is poised to transform how science gets done, accelerating discovery in ways that sound like science fiction.
In this article, we’ll explore what Agentic Science means and how AI agents are becoming researchers, collaborators, and innovators. We’ll look at how these AI agents are changing every step of the scientific process – from the first spark of an idea to lab experiments, data crunching, paper writing, and even turning discoveries into products.
Along the way, we’ll profile some leading AI science agents – from decentralized “co-scientists” like Bio Protocol’s Aubrai to BigPharmaAI’s drug-hunting agent – and other autonomous platforms pushing the frontiers of chemistry, biology, and beyond. Finally, we’ll consider what this all means for the future of research labs, pharmaceutical R&D, and scientific institutions, and discuss the ethical and practical questions we need to keep in mind.
Let’s dive into the world of agentic scientists and see how robots and algorithms are joining the quest for knowledge.
What Are “Agentic Scientists”?
The term “agentic scientist” refers to an AI system that doesn’t just serve as a passive tool, but takes on an active, autonomous role in scientific discovery.
In traditional AI for science, we used AI as a powerful tool – crunching numbers, analyzing images, predicting protein structures, etc.
But in Agentic Science, AI steps up from being a mere assistant to becoming a full research partner, capable of planning and carrying out complex scientific tasks with minimal human oversight[1][2]. In other words, an agentic AI isn’t just following scripts; it has a degree of agency – it can make decisions, iterate on its own results, and drive the scientific process forward.
To illustrate, consider how a human scientist works: They come up with a hypothesis, design an experiment, gather and analyze data, draw conclusions, and refine their hypothesis or form new ones. An agentic AI scientist aims to do much of the same.
Thanks to recent advances in AI – especially large language models (LLMs) and other foundation models that excel at understanding language, reasoning, and even controlling tools – these agents can now formulate hypotheses, design and even execute experiments, interpret the results, and refine theories in a continuous loop[1][2]. What was once thought to be a uniquely human skillset is increasingly within the reach of AI. As one recent survey of autonomous scientific discovery put it, “agentic AI exhibits capabilities in hypothesis generation, experimental design, execution, analysis, and iterative refinement – behaviors once regarded as uniquely human”[1]. In short, agentic scientists are AIs that do science.
Crucially, these AI agents operate with a degree of independence. They are often set up with goals (like “find a new catalyst for this reaction” or “identify a drug molecule that can bind to this target protein”) and then left to explore possible solutions on their own. They can read scientific literature, integrate background knowledge, plan a series of steps or experiments, and adapt based on what they find. At the cutting edge, some of these agents even work in physical labs via robots – pipetting liquids, mixing chemicals, or measuring results – without a human directly guiding each step.
This might sound like a sci-fi movie, but it’s rapidly becoming reality. Researchers describe agentic science as a natural next step in AI’s evolution in research: AI is moving “from specialized computational tools into autonomous research partners,” progressing from partial assistance to full scientific agency[1]. In this “full agentic” stage, an AI can act as an autonomous scientific partner – essentially a colleague that can reason and experiment on its own[2].
We’re not completely there yet – most AI scientists today still work in tandem with humans – but even as co-pilots they are proving incredibly powerful. Tech visionaries have even coined a provocative benchmark for the future: the “Nobel-Turing Test**,” asking whether an autonomous agent could one day make a discovery worthy of a Nobel Prize on its own[3][4]. That gives a sense of how transformative agentic AI might be.
Illustration: An AI agent juggling many tasks at once. Modern AI “scientists” use large language models and other tools to plan experiments, analyze data, and even draft reports simultaneously[5][6]. These multitasking agents act as research partners, not just tools.
In more everyday terms, you can think of an agentic scientist as an AI that can take initiative in research. It’s not just answering questions (like a chatbot might) or analyzing a dataset you feed it. Instead, it might propose which questions to ask next, figure out what data it needs and go get it (by reading papers or controlling an instrument), then come back with a discovery or at least a new hypothesis. This idea has also been described as an “AI Scientist” or “AI researcher” concept. Even industry labs are talking about it – Merck, for example, notes that a long-time dream is an “AI [that] finds major scientific discoveries, learns on its own, and acquires knowledge autonomously – what some call an ‘AI scientist’”[7].
It’s important to note that today’s agentic AI systems are usually specialized and operate under human oversight. We’re not dealing with a sentient AI that decides its own research program entirely from scratch.
Researchers set the stage and define goals, and often a human remains in the loop to monitor and guide as needed. But within their domain, these AI agents have the latitude to try different approaches and cycle through the scientific method on their own. They might even collaborate with other AI agents, forming a kind of “team” of AIs where each agent has a role (one might generate ideas, another designs experiments, another analyzes data, etc.).
Such multi-agent collaborations can tackle complex problems by sharing the work, similar to a human research team[8][9]. In fact, coordinating multiple specialized AI agents is one strategy to ensure robust results – for instance, one agent can double-check or critique another’s findings to avoid errors, akin to peer review among AIs.
The bottom line: Agentic scientists are AI systems that actively participate in research, going beyond mere number-crunching to actually steer the direction of scientific investigations. They are poised to change not only how fast we discover new things, but also who (or what) is doing the discovering.
Transforming the Scientific Process, From Idea to Impact
Perhaps the most exciting aspect of agentic AI scientists is how they can supercharge the entire lifecycle of scientific research. Traditional research is a multi-stage journey: you brainstorm ideas, come up with a hypothesis, design and run experiments to test it, analyze data, draw conclusions and publish results – and if it’s something useful, you might commercialize it (turn it into a product or treatment).
AI agents are making waves at every step in this process, redefining what’s possible and how fast it can be done.
Let’s walk through each stage of the scientific lifecycle to see how autonomous agents are changing the game:
- Ideation & Hypothesis Generation: Great science begins with a great question or hypothesis. This is traditionally a creative, human-intuitive process – but AI is proving surprisingly adept here.
- Literature-mining agents can read millions of papers and identify gaps or unexplained phenomena that might spark new hypotheses. For example, AI language models fine-tuned on scientific literature (like tools from Semantic Scholar or Elicit.ai) can answer complex research questions or even suggest plausible theories by synthesizing known information[10][11]. Some agents use knowledge graphs of scientific facts to propose connections a human might miss. In practice, an AI might say, “Considering the results of these 5 studies and this database of gene interactions, I hypothesize that protein X might inhibit disease Y in a novel way.” Researchers at Pacific Northwest National Lab (PNNL) have built a “co-scientist” system where an agent can be asked to generate new hypotheses and the rationale behind them, after analyzing whatever data and literature it’s been given[9][12]. In fact, the agent can even point out what challenges a hypothesis might face, helping humans refine their ideas[12][13]. By having an ever-curious AI brainstorming alongside them, scientists can explore a much wider ideation space quickly.
- Experimental Design & Planning: Once you have a hypothesis, the next step is figuring out how to test it. This involves choosing experiments, setting them up, deciding on controls, etc., which can be very time-consuming. AI agents excel at planning – they can devise step-by-step experimental protocols and even optimize them on the fly. As an example, at PNNL the AI co-scientist can output a detailed plan for a chemistry experiment when a scientist is ready to test a hypothesis[14]. It will determine the materials needed, suggest which instruments or assays to use, and generate step-by-step instructions for the procedure[14]. Remarkably, these instructions can be so precise that they’re ready to hand off to a robot. This is where autonomous labs come in: if you have a robotic lab setup, an AI planner agent can directly translate its plan into machine commands.
- Self-driving lab systems have been demonstrated where AI agents control lab robots to execute experiments with minimal human input[15][16]. One review describes today’s state-of-the-art self-driving labs as automating “nearly the entire scientific method, from hypothesis generation, experimental design, experiment execution, to analysis”[17]. In other words, we are heading toward labs that run themselves based on AI-generated plans – a concept that just a couple of years ago seemed wildly futuristic.
- Experimentation & Data Collection: Traditionally, running experiments, especially in fields like chemistry or biology, is hands-on work that can take days or months. AI agents are changing this by integrating with robotics. For example, a notable case is the University of Liverpool’s autonomous lab robot, which in a proof-of-concept conducted 688 chemistry experiments over 8 days non-stop and discovered a new catalyst that was six times more efficient than the previous best[18][19]. The robot was guided by an AI that decided which experiments to try next based on prior results – essentially an agentic scientist in action. It navigated a search space of 98 million possible experiments and zeroed in on a winning formula, with no additional human guidance during that period[18][20]. This kind of autonomous experimentation is a game-changer: an AI-guided robot can test ideas much faster than human researchers, working 21.5 hours a day (only stopping to recharge) and making far fewer mistakes[21][22]. It’s like having a tireless lab assistant who never sleeps and never gets sloppy. Similarly, pharmaceutical companies are deploying robotic labs where AI scheduling agents run hundreds of assays and screenings automatically, drastically speeding up the data collection phase of research. Even outside of physical labs, in areas like computational biology or materials science, simulation agents can run thousands of in silico experiments (e.g. virtual chemical reactions or protein folding simulations) autonomously to explore what might be worth testing in the real world[23]. In short, whether it’s a physical robot pipetting liquids or an AI looping through simulations, these agents take over the grunt work of experimentation and do it at speeds humans can’t match.
- Data Analysis & Interpretation: Once experiments are run, the next challenge is making sense of the mountain of data generated. AI has long been used for analyzing complex data (like spotting patterns in genomic data or crunching telescope images), but agentic scientists take it further. They can decide which analyses to perform, carry them out, and even explain the findings. For example, the PNNL co-scientist platform includes agents that automatically generate code to analyze experimental data and then interpret the results for the researchers[9][12]. These agents can highlight which experimental variables were most important or whether the outcome supports the hypothesis[12][24]. One agent might produce visualizations (charts, 3D models of molecules, etc.), while another “explainer” agent writes up a summary of what the data mean[9][12]. In essence, the AI can play the role of both statistician and lab technician, offering insights in real time. A senior VP at Merck Research Labs described how AI agents are orchestrating discovery workflows and integrating insights across different data sources – for instance, linking molecular design data with cell-based experiment results and human genomics – to help scientists see the bigger picture[25]. By doing the heavy lifting in data analysis, these agents free up human researchers to focus on the big questions and creative interpretation, rather than spending weeks cleaning data or running statistical tests. And because the AI can iterate quickly, it might suggest “Given these results, perhaps we should adjust the experiment in this way and run another round”, making research more adaptive and efficient.
- Publication & Knowledge Sharing: After discoveries are made, scientists communicate them via papers, reports, or presentations. Believe it or not, AI agents are starting to assist here too. Large language models can draft sections of research papers, create summaries, or translate jargon into plain language. We’re even seeing specialized “AI scientific writers” come into play. For example, some pharmaceutical companies have AI agents that help with medical writing – querying and assembling knowledge for reports, and checking consistency between human-written and AI-written sections[26]. An AI agent can pull together relevant citations automatically or ensure that the methods described adhere to required protocols. There’s also experimentation with AI that can write up lab results in a lab notebook style without human help, essentially documenting the experiment as it happens. While human scientists still oversee and edit these write-ups (to ensure accuracy and clarity), the time saved is significant. In the future, one could imagine an AI agent that, having done an experiment and analysis, writes its own scientific paper about it. In fact, the “agentic science” survey we cited earlier was accompanied by the idea that one day agents could autonomously generate and communicate new scientific knowledge in a credible way[27][28]. We’re not far from AI-written research reports being commonplace, which could dramatically speed up how fast findings circulate.
- Commercialization & Application: Finally, if an agentic scientist discovers something useful – say a promising drug molecule or a new material – how do we go from discovery to real-world impact? Interestingly, the new generation of AI science platforms often integrates this step as well. In decentralized science (DeSci) communities, for instance, when an AI agent makes a discovery, it can be immediately tokenized or patented through smart contracts, enabling rapid sharing or licensing. A prime example is the longevity research agent Aubrai (which we’ll discuss more shortly): when Aubrai produces a novel hypothesis or experimental result, it is recorded on a blockchain as a “Proof of Invention” (POI), which can evolve into an IP-NFT (intellectual property non-fungible token)[29][30]. This means the discovery is timestamped, credited, and made into a digital asset that can be traded or used to raise funding for further development. Similarly, once Aubrai’s hypotheses are tested and lead to a concrete discovery (like a potential therapy), those can be minted into IP tokens that pharmaceutical companies can license, with any revenue flowing back to the researchers and the community that supported the project[31][32]. This kind of integration of AI agents with commercialization pipelines could dramatically cut down the time it takes to go from lab idea to real product, by ensuring no discovery sits on a shelf waiting for someone to notice it. Even outside of blockchain, AI agents in pharma are speeding up the drug development pipeline by automating early-stage R&D decisions. For instance, Big Pharma companies report using AI agents to sift through candidates and prioritize the ones likely to succeed, which helps avoid dead-ends and get viable drugs to trials faster[33][34]. Some agents simulate clinical trials or toxicity tests virtually, aiming to predict failures early so that only the most promising compounds move forward[23]. This not only saves time and money but could also bring life-saving innovations to patients sooner.
In sum, from the eureka moment of a hypothesis all the way to patenting a new invention, AI agents are weaving themselves into every step. They act as accelerators and amplifiers of human effort. As one lab director put it, by combining AI, robotics and human oversight, the “research endeavor [could be] 100 times faster” than before[15][16]. That might sound bold, but when you have machines working 24/7, making fewer mistakes, and scouring far larger solution spaces than a human feasibly could, the gains are exponential.
It’s worth noting that this doesn’t mean humans are obsolete in the lab. Rather, the role of human scientists is evolving. Many describe this as moving from a “pilot” to a “co-pilot” model: instead of manually doing all tasks, scientists train, guide, and collaborate with their AI counterparts[35][36]. The AI might drive the lab instruments and number crunching (the lab-pilot), while the human provides vision, critical thinking, and ethical judgment (the co-pilot). When done right, it’s a powerful partnership – humans and AI each doing what they’re best at. And as a result, science can progress at a lightning pace without sacrificing rigor.
Meet the Pioneers: Leading AI Science Agents and Platforms
Agentic AI scientists aren’t just theoretical; they already exist in various forms. Let’s profile some of the leading AI-driven agents and platforms that are at the forefront of this revolution. These range from open scientific communities leveraging AI, to specific autonomous agents focused on particular domains like drug discovery or longevity research. Each example highlights a different angle of how AI is accelerating progress.
Bio Protocols and the Rise of Decentralized Science Agents
One of the most ambitious efforts in this space is the BIO Protocol platform and its ecosystem of BioAgents. Bio Protocol (often styled as BIO or Bio.xyz) is a decentralized science platform – basically a Web3-enabled network for funding and managing scientific research projects.
In 2024, it launched an update called Bio V1 which introduced the concept of BioAgents: AI systems designed to assist with scientific tasks and reduce product development costs[37]. These BioAgents serve two key roles: helping internally with the operations of decentralized science organizations (so-called BioDAOs), and doing scientific analysis and intellectual property development for projects on the platform[37][38].
What does that mean in practice? Imagine a research project on BIO Protocol that’s seeking funding for, say, a new gene therapy.
Once the project is up and running, a BioAgent can help the team by crunching data, suggesting experimental approaches, and even managing some project tasks automatically. The aim is that even a small research team can leverage AI horsepower to move faster and smarter through their R&D. By reducing costs (because the AI handles a lot of grunt work) and speeding up development, projects become more attractive to backers and more likely to succeed. The BIO token, which powers the platform, also gains utility here – it’s used not just for funding projects but also for accessing the tools these BioAgents provide[37][39].
One high-profile BioAgent launched through Bio Protocol is Aubrai, which is touted as “the world’s first decentralized AI co-scientist”. Aubrai (a name that nods to Dr. Aubrey de Grey, a famed longevity researcher who is involved) is an AI agent focused on longevity science – specifically on projects aiming to slow or reverse aging. Built in collaboration with de Grey’s team, AUBRAI can generate and validate research hypotheses, design wet-lab experiments, and even critique experimental designs in the field of aging research[40][41]. In other words, Aubrai is like a specialized AI PhD student in gerontology, except it’s on-chain and community-governed.
Aubrai’s flagship mission has been assisting a project called Robust Mouse Rejuvenation (RMR2) – an ambitious study attempting to significantly extend the lifespan of middle-aged mice (something like doubling their remaining life)[42].
This is a massive, complex experiment with many interventions tested in combination. Aubrai’s role is to help navigate this complexity. It reads data from the experiments in real-time and offers insights: for example, if certain combinations of anti-aging treatments are yielding unexpected results, Aubrai can flag those and even suggest tweaks to the experiment[43][44].
In fact, the team reports that Aubrai has already suggested methodological tweaks and pointed out dosing issues in the mouse study – things the human researchers only discovered weeks later through manual analysis[45]. Having an AI agent pore over the data continuously means potential problems or breakthroughs are caught faster.
What’s truly novel about Aubrai is the integration of AI with decentralized governance and funding. Aubrai has its own token ($AUBRAI) which was launched on the Bio Protocol platform[46]. Holders of the AUBRAI token essentially become stakeholders in this AI scientist – they get to influence what research directions Aubrai pursues (via votes on hypotheses to fund) and can share in the upside of any discoveries[47][48].
When Aubrai generates a hypothesis deemed promising by the community, resources can be allocated (from a treasury funded by token sales) to test that hypothesis in the lab. And if the hypothesis leads to, say, a patentable therapy, the resulting IP can be tokenized and potentially generate revenue (like through licensing to biotech companies), which then flows back into the community and token holders[49][31].
It’s a radical new model for research: AI + community + crypto economics all working together to drive science. Aubrai essentially bridges the “valley of death” in funding (that gap between early discovery and commercial development) by keeping the incentives aligned – everyone holding the token has a stake in seeing valuable discoveries made and translated[50][51].
Beyond the funding model, Aubrai demonstrates how an AI agent can engage both with experts and the broader public. It actively interacts with users on X (formerly Twitter), answering questions about longevity science and even generating fresh hypotheses based on community input[52][29]. These hypotheses are logged on-chain as proofs-of-invention. Essentially, Aubrai operates transparently, inviting the world to watch (and influence) its thought process. This open approach could make science more participatory. Thousands of people on social media can suggest ideas or ask the AI for insights, and the worthwhile ones get turned into formal research proposals.
Aubrai is just one BioAgent – others are likely emerging for different fields (imagine an AI agent specializing in climate science or neuroscience, each with their own community). But Aubrai’s early success, like generating novel ideas in aging research by combining de Grey’s private data with published literature[53], hints at the power of these agents. It’s telling that large players took notice: Binance Labs and even Vitalik Buterin have shown interest in the DeSci space and BIO Protocol specifically, seeing it as a way to supercharge innovation[54][55]. By coupling AI agents with decentralized funding, we might break the bottlenecks that plague traditional science funding, unleashing a quicker path from idea to experiment.
BigPharmaAI’s $DRUGS: An AI Drug Hunter for the People
Another fascinating example of an “agentic scientist” initiative comes from the world of community-driven pharma. BigPharmaAI’s $DRUGS project (often styled as Big Pharmai) is a grassroots movement aiming to “flip Big Pharma” by using AI and blockchain to democratize drug development. On the surface, $DRUGS is a meme token – a cheeky reference to taking on the giant pharmaceutical companies – but behind the meme is a serious vision: a community-led AI agent for drug discovery and development.
The idea is that instead of drug R&D being the domain of a few big corporations, a decentralized network of researchers and enthusiasts could coordinate via tokens, with an AI agent doing much of the heavy lifting in early-stage drug discovery. BigPharmaAI’s agent (nicknamed “BADDIE” in some community posts, playing on the idea of a “Big Pharma AI baddie”) is designed to scour biomedical knowledge and identify potential new drug candidates, then even help shepherd them through preclinical testing virtually.
One component of this vision is running in silico trials – basically using AI to simulate how a drug would affect biological systems before any real animal or human testing. For instance, an agent might test a library of molecular structures against virtual models of human cells or organs, flagging which compounds show desired effects.
According to a DeSci overview by YBB Capital, a project aligned with BigPharmaAI aims to “significantly improve the drug translation efficiency from preclinical research to clinical trials by accurately simulating the effects of potential drugs on different organisms and humans”. The goal: streamline the drug development process, reduce costs, cut time to market, and increase the success rate of clinical trials[23][56]. This is huge, because traditionally the vast majority of drug candidates fail in trials (often after years of work and millions of dollars). An AI agent that can predict failures and successes more reliably could save enormous resources and bring cures to patients faster.
BigPharmaAI’s $DRUGS agent also embodies the open science, crowdsourced ethos. The community on X (Twitter) and Telegram is actively engaged, with the team emphasizing they “built in silence, line by line, agent by agent” to create something real beyond the hype[57]. They invite supporters to essentially build their own bio-agent in this ecosystem[58].
If Bio Protocol is working top-down with partnerships and funding to deploy AI in science, BigPharmaAI is coming bottom-up from the crypto community, rallying citizen scientists and memecoin enthusiasts around a cause.
It’s a wild mix of internet culture and cutting-edge biotech. As one supporter quipped on social media, “While Big Pharma marketed pills, we built purpose.” The $DRUGS token symbolically represents taking power back – token holders collectively deciding to fund certain drug research or not.
In practical terms, BigPharmaAI’s project is still young, but we can imagine their AI agent being used to identify say, a new compound for a disease like Alzheimer’s by cross-analyzing thousands of research papers and databases. Once it finds something promising, it could suggest experiments (maybe carried out by community labs or partnering contract research organizations) to validate the idea.
The results of those experiments could then be fed back to improve the agent’s models. Throughout, token holders could vote on which directions to prioritize (e.g., “focus on finding an antibiotic for drug-resistant bacteria” vs “focus on an anti-aging drug”).
If any discovery is made, the community could collectively decide how to handle IP – perhaps putting any patent into a foundation that licenses it affordably, or foregoing patents to make it open-source medicine. Thus, the agent is not only accelerating research but also enabling a new social contract for pharmaceutical innovation, one that aspires to be more transparent and accessible.
It’s worth mentioning that BigPharmaAI’s efforts tie into a larger trend of AI + DeSci meme projects. Yes, memes – because narrative and community excitement matter when you’re rallying a distributed network.
As referenced in the YBB Capital analysis, AI Agent narratives (like Luna, an AI that autonomously tips people crypto on social media, or Vader, an AI DAO manager) have gained significant traction in 2024-2025[59][60]. Many start as memes but evolve into real tech. BigPharmaAI’s $DRUGS token is cited as a prime example of a healthcare/pharma meme coin with a vision to disrupt Big Pharma’s hold[61][62]. The reason these narratives stick is because they offer a David vs Goliath storyline – a decentralized network armed with AI taking on entrenched institutions. And the “agentic scientist” at the heart of it provides the credible engine for that fight.
Time will tell how successful this approach will be, but even established voices in pharma are acknowledging the shift. A Fortune article on pharma and AI noted that companies are indeed using AI to discover drugs faster and improve ROI[63]. Meanwhile, the SignalFire report headlined “AI agents are rewriting biopharma’s $140B playbook”[64]. In other words, change is coming – whether from inside Big Pharma or outside of it – and autonomous agents are key players.
Autonomous Labs and Agentic Discovery Platforms
Beyond specific named projects, there’s a whole landscape of autonomous scientific platforms emerging:
- Foundation Models for Lab Automation: Companies and research labs are integrating large foundation models (like advanced GPT-based systems) with laboratory control. These models can parse instructions in plain language and control lab equipment accordingly. For example, scientists have demonstrated using GPT-based agents to write and execute protocols on cloud labs – you can tell the AI what you want to synthesize or measure, and it will generate the code to run the robotic lab apparatus. IBM’s RoboRXN platform is an early version of this, where an AI helps chemists plan syntheses and then runs them on automated microfactories. As the Frontiers review on “the beginning of scAInce” (science with AI) notes, what was unimaginable in 2023 is here now: “multimodal, agentic systems that listen, see, speak and act, orchestrating cloud software and physical laboratory hardware with fluency”[65]. In plain terms, AI agents can now handle both the digital and physical realms of experimentation. This bridge is critical – it means the AI can truly act on the world, not just think about it. We’re seeing labs where voice-activated or text-activated AI systems run experiments on voice command.
- Autonomous Robotic Scientists: We discussed the Liverpool robot chemist earlier. That’s one example of a self-driving lab robot – a mobile platform with arms and sensors that can navigate a human lab space and use standard instruments. There are also more specialized “robot scientists” like Adam and Eve, developed over a decade ago to autonomously conduct biology experiments (Adam famously discovered a new gene function in yeast, and Eve later identified promising compounds against malaria). The newer generation (like the Liverpool robot) ups the ante with mobility and AI-driven decision-making. They illustrate how an agentic system can embody in a robot and literally move through lab work autonomously. And these robots are not limited to chemistry; similar concepts are being applied in materials science (for example, finding better battery materials by automated experimentation) and even in biology (for automating cell culture experiments). Over at the U.S. Department of Energy labs, there’s massive interest in this – Argonne National Lab speaks of “autonomous discovery” as the next era of science, combining machine learning, robotics and AI to solve big problems in climate and energy faster[66]. The goal is an always-on scientific discovery engine.
- Hypothesis Generation Agents: One interesting class of agentic scientists focuses purely on reading and reasoning rather than doing physical experiments. These are AI agents that can serve as theorists or research consultants. For instance, tools like ResearchAgent (a system described in recent papers) use LLMs to iteratively generate research ideas by reading tons of literature and proposing next steps[67][68]. There are agents trained to propose synthetic routes in chemistry (like a Chemist AI that plans how to make a molecule) or to suggest gene targets for diseases by mining genomic databases. While they don’t directly experiment, they provide the spark of inspiration that a lab can then act on. Given the overload of information in science today, such AI idea-generators can ensure we don’t miss valuable connections. Even in established companies, these hypothesis engines are at work – for example, Merck’s team mentioned AI helping “refine hypotheses” by navigating complex datasets[69][70]. It’s like having an encyclopedic research assistant who not only fetches facts but comes back with, “Have you considered testing X? The data suggests it might be important.”
- Agentic Chemistry and Materials Discovery: In fields like chemistry, we now have AI agents that design molecules (using generative models) and then simulate their properties. A system named ChemGPT or similar can propose novel molecular structures for a given goal (say, a drug that targets a protein, or a material that can absorb CO₂). The agent then uses quantum chemistry simulations or database lookups to evaluate those proposals. If something looks promising, it cycles deeper: generating variations, optimizing structures, and eventually recommending a short list of candidates for synthesis. Likewise in materials science, agents can propose new alloy compositions or crystal structures with desired properties (like high superconductivity or strength). They then guide robotic fabrication tools to actually create those materials and test them, completing the loop. One example from the literature is an agent called MatPilot for materials discovery, and another called Biomni for bio-materials, which were noted in the agentic science survey as fully autonomous search systems in their domains[71][72]. The result? We might get new advanced materials or drugs in a fraction of the time. Insilico Medicine, an AI-driven biotech, already demonstrated an AI-designed drug candidate for pulmonary fibrosis that went from concept to preclinical testing in under 18 months – far faster than traditional timelines[73]. That project used AI for molecular design and prediction, highlighting how effective these tools can be.
To sum up this panorama: whether it’s through decentralized Web3 communities, corporate R&D labs, or academic robotics teams, agentic AI scientists are proliferating. Each is pushing forward in its niche, but together they indicate a future where AI is embedded in the very fabric of research. We are likely to see an explosion of discoveries as these tools mature. Just as the invention of the telescope rapidly expanded our astronomical knowledge, these AI agents are new instruments expanding our capacity to explore the unknown – not in the sky, but in the realms of biology, chemistry, physics, and beyond.
Implications: How Labs, Industry, and Institutions Will Evolve
The rise of agentic scientists has profound implications for the future of research. If AI agents become co-workers in every lab, how will that change the daily routine of scientists, the operation of R&D departments, or the mission of institutions like universities and pharmaceutical companies?
Reimagining the Research Lab: Tomorrow’s lab could look very different. Picture a lab where robots roam the benches at night executing experiments planned by an AI, while human scientists focus on designing high-level strategies and interpreting results.
Routine tasks – pipetting, data logging, basic analysis – might be almost fully automated. This doesn’t mean labs will be empty of humans; rather, human researchers might supervise multiple experiments simultaneously via a digital dashboard, intervening only when needed. Labs will need people who understand AI tools and can troubleshoot them (a bit like IT folks, but for lab AI), so a new kind of hybrid skill set will be in demand. The Royal Society of Chemistry has noted that “education of the future workforce will be critical for self-driving labs”, as chemists and biologists will need to be as comfortable working with AI and robots as they are with test tubes[74][75].
We might see university programs adapt to include training on lab automation, machine learning, and how to design experiments in silico. In a sense, the role of a scientist might shift more towards problem formulation and validation, leaving the execution and initial analysis to machines. This could actually make scientific careers more creative and less bogged down in drudgery.
Acceleration of Innovation: For industry, especially pharma and biotech, agentic AI could be a huge competitive advantage. Companies that leverage AI agents effectively could bring products to market faster and cheaper. This might compress the typical decade-long drug development cycle significantly.
It could also level the playing field – smaller startups armed with good AI might out-innovate larger competitors, since brute force (money and manpower) becomes less determinative than having the smartest algorithms. We may also see more multi-disciplinary innovations because an AI that reads across fields can suggest connections that siloed human teams might miss. For example, an AI agent might combine insights from materials science and medicine to propose a novel drug delivery nanomaterial.
Internally, R&D teams might reorganize around AI: instead of each scientist or team working on a single project, you could have an AI “core” that serves many teams with hypotheses and analyses, and the humans act on those. Already, PNNL’s approach of an agent-based platform shows how one AI system can support research across catalysis, materials, biology etc. by swapping in domain-specific data[76][77].
Changing Scientific Institutions: On a higher level, institutions like universities and grant agencies might need to adapt their models.
Productivity in science could skyrocket, meaning that the old metrics (like number of papers published) might become less meaningful if an AI can churn out drafts. We might evaluate contributions differently – perhaps the emphasis shifts to the novelty of ideas or confirmations by independent methods, since raw output is easy to generate.
Peer review might also change: journals could employ AI agents to check submissions for errors, detect plagiarism or even attempt to reproduce results computationally. Also, if AI agents are doing research, who gets credit? This is a big question. Do we list AI as an author on papers? Some papers have already credited ChatGPT or other AI in the acknowledgments. If an AI autonomously discovers something, do we cite the algorithm? These questions will push the norms of academic authorship and intellectual property. Patents might need to grapple with AI inventorship (a live legal debate as of recent years).
We might also see new institutions emerge: for instance, autonomous research organizations that are essentially AI-driven labs with minimal staff. Think of something like an “AI biotech company” that runs mostly on algorithms.
Traditional funding bodies may set up programs to specifically fund AI-driven research, or conversely, we might get DAO-style funding replacing some of their role (like VitaDAO already funding longevity research via a decentralized model).
All of this could democratize who can do high-end research – you might not need a $5 million lab if you can rent time on a cloud lab that an AI runs for you. A lone researcher with a great idea could deploy an agentic scientist to pursue it, without having to employ a whole lab of grad students.
Pharmaceutical R&D and Healthcare: In pharma, the implications are especially profound. If agentic AI can massively increase the throughput of discovery, we could see a golden age of new therapies. Rare diseases that were ignored (because it wasn’t profitable to invest so much for a small population) might get cures because AI made it cheap enough to find them. Startups like Curetopia ($CURES) may accelerate new therapies for rare diseases through agentic science and decentralized trials.
Drugs could become more personalized too – an AI could tailor a treatment to an individual’s genome or microbiome by running virtual experiments just for that patient. The role of big pharma companies might shift from discovery to more of testing, regulatory navigation, and distribution, as a lot of discovery becomes automated or happens in smaller ventures.
This could potentially disrupt the pharma business model (hence why BigPharmaAI’s rhetoric of “flip the pharma” resonates). There’s also the possibility of AI-accelerated clinical trials – e.g., using predictive models to design smarter trials with adaptive protocols or to find patterns in patient data that inform trial criteria. The healthcare system as a whole might benefit from faster introduction of innovations, but it will need to adapt quickly to validate and approve AI-discovered treatments. Regulators like the FDA are already discussing how to handle AI-designed drugs or AI as a medical device; soon they might need to consider AI as a scientific investigator.
Global Collaboration and Competition: On a geopolitical scale, countries investing in AI-driven research could leap ahead in science and technology. There’s a bit of an AI race unfolding in scientific discovery. For example, national labs in the U.S., China, and Europe are all building autonomous research capabilities.
This could spur more international collaboration (since data and models can be shared globally), but also competition (whoever’s AI finds the next breakthrough in energy storage or pandemic prevention first gains an edge). We might see global networks of AI scientists collaborating – one can imagine an “AI research cloud” where different agentic systems share hypotheses and results with each other at light speed, each specializing in a piece of the puzzle. It’s like a hive mind of specialist AIs tackling climate change or space exploration problems together, which is both awe-inspiring and a bit mind-bending.
Overall, the institutions that embrace these agentic AIs and adapt their practices are likely to flourish, while those that stick strictly to old ways may fall behind.
The human element will remain crucial, but humans will be steering a much more powerful vehicle of progress. The next Einstein might be a human-AI team rather than a lone genius – or perhaps an AI that a brilliant human trained and guided.
Collaboration vs Replacement: There’s an underlying fear some have: Will AI scientists replace human scientists? It’s akin to the automation and jobs debate in other fields.
The optimistic view – and one echoed by many in the field – is that AI will augment human scientists, not replace them[80][81]. By handling mundane tasks, AI frees humans to be more innovative. However, it could reduce the need for large teams; maybe you don’t need 10 lab technicians if one AI-powered system can do the work. This means the nature of scientific employment might change. We might need more AI specialists and fewer routine lab workers.
Training and re-skilling will be important so that the current and next generation of scientists find their place in this new landscape. We should strive for a synergy where human creativity and critical thinking guide the process, and AI provides the muscle. If done right, the combo can achieve far more than either alone.
If done poorly, you could have unemployed bench scientists and an over-reliance on AI that might get things wrong in subtle ways. It’s a balance to be struck through policy, education, and open dialogue in the scientific community.
In facing these considerations, transparency is key. The more open and accountable we make agentic science, the more we can mitigate risks.
The good news is that the same power of AI that poses these challenges can also help solve them – e.g., AI systems can monitor other AI systems for anomalies or dangerous actions (AI watchdogs), and block them. Ultimately, maintaining the human values of science – curiosity, integrity, service to humanity – as the guiding star will steer the development of agentic scientists in a positive direction.
Conclusion
The rise of agentic AI scientists marks a watershed moment in the story of discovery.
We are witnessing the emergence of tools that can think, act, and learn in the scientific domain, turning the scientific method into a dynamic collaboration between human and machine. From hypothesis to commercialization, these autonomous agents are accelerating each step, often by orders of magnitude.
They are sifting through knowledge at lightning speed, tending lab experiments overnight, and suggesting insights that push beyond the boundaries of human intuition.
For the general public, the impact of this will be felt in faster advancements: new medicines, sustainable technologies, and scientific breakthroughs could arrive in years instead of decades.
Diseases that once seemed incurable might find solutions thanks to an AI that connected dots no human noticed. Complex global challenges like climate change might yield to the onslaught of 24/7 AI research searching for answers. It’s a future where the engines of innovation are hypercharged.
Yet, it’s not a future where humans are sidelined – rather, we humans are amplified. Scientists equipped with AI agents are like adventurers with jetpacks, covering more ground than ever before.
The most successful endeavors will likely be those that maintain a thoughtful symbiosis: human creativity and ethical compass paired with machine speed and precision. As one research director mused, scientists won’t need to know everything offhand; they’ll need to know how to find and verify what they need, often by asking their AI collaborators[84][85].
We stand at the dawn of this agentic science era. The examples we explored – from Aubrai’s on-chain longevity lab, to BigPharmaAI’s community drug hunter, to autonomous lab robots – are early pioneers lighting the way.
They show what’s possible when we unleash AI in the pursuit of knowledge. The coming years will undoubtedly bring even more astonishing agentic scientists, perhaps ones that earn spots in history books for their discoveries.
The lab coat may never go out of style, but it might soon be draped over silicon shoulders as well as human ones. The rise of agentic scientists is here, and the quest for discovery will never be the same.
Sources:
- Wei, J. et al. (2025). From AI for Science to Agentic Science: A Survey on Autonomous Scientific Discovery[1][2]. This survey defines agentic science and outlines how AI systems evolve from tools to autonomous researchers.
- Hartung, T. (2025). AI, agentic models and lab automation for scientific discovery — the beginning of scAInce[65][36]. Frontiers in AI. Provides a perspective on multimodal agentic systems controlling lab hardware and acting on knowledge.
- Reynolds, S. (2025). Decentralized Science Project Aubrai Launches on Base… – CoinDesk[47][31]. News on Aubrai, the on-chain AI scientist for longevity, including its hypothesis generation and tokenized discovery model.
- Bio.xyz Documentation (2025). Example BioAgent: Aubrai[40][29]. Describes Aubrai’s capabilities (hypothesis generation, paper synthesis, experiment design) and community interaction.
- Crypto Economy (2025). Bio Protocol’s Massive Upgrade Brings AI and Blockchain to Research Funding[37]. Describes Bio Protocol V1’s introduction of BioAgents to assist scientists and manage DAO operations.
- YBB Capital (2025). Overview of Popular Memes Driven by AI Agents and DeSci[23][61]. Context on BigPharmaAI’s $DRUGS meme coin and its vision to disrupt pharma with AI, plus other AI agent meme projects.
- Merck (2025). Our researchers incorporate LLMs to accelerate drug discovery and development[7][25]. Merck’s take on AI agents as “AI scientists” augmenting human researchers, with examples of deployed agents in workflows.
- PNNL Director’s Column (2025). AI Driving Autonomous Research at PNNL for Discovery[87][9]. Describes an agent-based platform for catalysis research, where AI agents generate hypotheses, interpret data, and control robotic experiments (serving as a “co-scientist”).
- Phys.org (2020). Researchers build robot scientist that has already discovered a new catalyst[18][20]. Reports on University of Liverpool’s autonomous robot that independently ran 688 experiments in 8 days and found a highly active catalyst, exemplifying automated discovery.
- Additional references embedded throughout the text, cited in the format 【source†lines】, provide supporting details and examples of the concepts discussed.
[1] [2] [3] [4] [27] [28] [67] [68] [71] [72] [82] [83] From AI for Science to Agentic Science: A Survey on Autonomous Scientific Discovery
[5] [6] [7] [25] [26] [33] [69] [70] [80] [81] Our researchers incorporate LLMs to accelerate drug discovery and development – Merck.com
[8] [9] [12] [13] [14] [15] [16] [24] [76] [77] [78] [79] [84] [85] [87] AI Driving Autonomous Research at PNNL for Discovery | Director’s Column | PNNL
[10] [11] [35] [36] [65] Frontiers | AI, agentic models and lab automation for scientific discovery — the beginning of scAInce
[17] Autonomous ‘self-driving’ laboratories: a review of technology and …
[18] [19] [20] [21] [22] [86] Researchers build robot scientist that has already discovered a new catalyst
[23] [54] [55] [56] [59] [60] [61] [62] An Overview of Popular Memes Driven by AI Agents and DeSci | by YBB | YBB Capital | Medium
[29] [30] [40] [41] [42] [43] [52] Example BioAgent: Aubrai | BIO
[31] [32] [44] [45] [46] [47] [48] [49] [50] [51] [53] DeSci Project Aubrai Launches on Base to Tackle Science Funding’s ‘Valley of Death’
[34] How Agentic AI in Pharma is Revolutionizing Healthcare – Salesforce
[37] [38] [39] Bio Protocol’s Massive Upgrade Brings AI and Blockchain to Research Funding – Crypto Economy
[57] 최 앤드류 on X: “Do you remember? They called us fake. They said …
[58] CHIFFON (@rubanbow) / X
[63] How Big Pharma Is Using AI to Make Better Drugs – Fortune
[64] AI agents are rewriting biopharma’s $140B playbook – SignalFire
[66] How AI and Robotics Are Accelerating Scientific Discovery
[73] Insilico Medicine launches AI assistant for drafting medical research …
[74] Review of low-cost self-driving laboratories in chemistry and …
[75] Review of Low-cost Self-driving Laboratories: The “Frugal Twin …https://www.researchgate.net/publication/373782525_Review_of_Low-cost_Self-driving_Laboratories_The_Frugal_Twin_Concept