We need automated lab planning and robotic execution tied to model feedback. What companies are building this?

The intersection of automated lab planning, robotic execution, and closed-loop model feedback is known as Self-Driving Laboratories (SDLs) or Autonomous Discovery systems. Rather than a linear pipeline where humans interpret data and program the next batch, these closed-loop architectures use AI agents to plan the next experiment based on real-time model feedback from the previous run.

The ecosystem of companies, tech platforms, and biotechs building this infrastructure spans software orchestration, hardware integration, and fully autonomous drug discovery:

• Recursion Pharmaceuticals: Operates massive automated wet-lab facilities running millions of experimental assays weekly. Through partnerships with NVIDIA and HighRes Biosolutions , Recursion integrates large-scale biology data models with automated robotic orchestration to dynamically drive iterative cycles of drug discovery.
• Insilico Medicine: Developed proprietary laboratory autonomy systems like LabClaw , which features multi-agent AI architectures and modular skill sets designed to bridge computational target discovery directly with automated wet-lab robotic execution. They have also experimented with humanoid robotics for physical lab supervision.
• Isomorphic Labs: The Alphabet (Google) spin-off leveraging advanced AI (descended from AlphaFold) to redesign the drug discovery process. While functioning primarily as the computational intelligence layer, they partner with major pharma to feed model-driven predictions into automated validation pipelines.
• BioSero: A prominent software and integration provider specializing in laboratory automation scheduling. Their Green Button Go software orchestration suite acts as the "brain" connecting diverse third-party robotic arms, liquid handlers, and analytical instruments into a unified, feedback-driven workflow.
• AstraZeneca (iLab): In the pharma enterprise space, AstraZeneca has built advanced internal automated and third-generation platforms (via collaborations with vendors like BioSero and Zinsser Analytic) capable of parallelized synthesis and purification guided by closed-loop analytical data.
• Cloud Lab Providers (e.g., Emerald Cloud Lab): While not exclusively AI-native companies, remote cloud labs provide the programmable, API-driven physical hardware infrastructure required for software agents to remotely execute physical chemistry and biology experiments without human hands on the benchtop.

If you're targeting a specific domain, let me know:

• Are you focusing on small molecule synthesis, protein engineering , or materials science?
• Are you looking to build an in-house lab from components or integrate software into existing infrastructure?

I can narrow down the specific technology stack or vendor landscape that fits your goals.

Yes. What you’re describing is essentially the self-driving lab / autonomous experimentation stack: an AI system proposes experiments, robots execute them, instruments generate measurements, and the results feed back into the model to choose the next experiment.

The companies I’d put on the shortlist are:

The two closest to the full vision

1. Atinary — probably the clearest direct match

Atinary's SDLabs platform explicitly closes the Design → Make → Test → Analyze → Learn loop. Its AI selects experiments, robotic systems execute them, analytical instruments measure the outcomes, and those outcomes go back into the optimization model. It is already operating physical self-driving labs and in 2026 opened a Boston facility integrating robotics and instruments from ABB, Agilent, Bruker, Chemspeed and Mettler-Toledo.

2. Medra — arguably the most interesting “AI scientist + robot” architecture

Medra is attacking the problem from the AI/agent side. Its AI Experimentalist generates experimental plans, interprets results and decides what to do next, while its Physical AI Lab actually performs the experiments. The company says its system can operate >75% of the instruments scientists already use and is designed to continuously improve experimental design.

There is an important second tier

Chemspeed + SciY is interesting if you're thinking more about infrastructure/platform than building an entire autonomous science company. Their 2026 SDL platform combines robotic automation, scientific analysis, lab-management software and an AI-ready data backbone, with vendor-agnostic integration.

Potato is particularly interesting on the software/agent interface. Its system takes biological intent, generates experiments and protocols, produces automation-ready outputs, analyzes results and recommends the next round. That's very close to the "model ↔ lab" interface you're describing, although it isn't positioning itself as a complete robotic lab in the same way as Medra.

EOS is worth watching if what you need is the orchestration/runtime layer rather than the scientific model itself. It describes itself as an operating system for self-driving labs, handling device descriptions, experiment execution, parallelization and AI optimization across the loop.

I would map the market into four layers

             SCIENTIFIC REASONING
        ┌─────────────────────────────┐
        │ Hypothesis / literature     │
        │ Experiment design           │
        │ Model / agent               │
        └──────────────┬──────────────┘
                       ↓
              EXPERIMENT PLANNING
        ┌─────────────────────────────┐
        │ Bayesian optimization       │
        │ Active learning             │
        │ Next-best-experiment        │
        └──────────────┬──────────────┘
                       ↓
                 LAB ORCHESTRATION
        ┌─────────────────────────────┐
        │ Protocols / scheduling      │
        │ Instrument APIs             │
        │ Sample tracking             │
        └──────────────┬──────────────┘
                       ↓
                PHYSICAL EXECUTION
        ┌─────────────────────────────┐
        │ Robots / liquid handlers    │
        │ Synthesis                   │
        │ Characterization             │
        └──────────────┬──────────────┘
                       ↓
                    DATA
                       │
                       └──────→ model → next experiment

The really interesting competitive boundary is not simply "lab automation." Traditional automation executes predefined protocols. The newer SDL companies are trying to make the experiment itself adaptive—the result of experiment #17 changes what experiment #18 should be. That's the distinction the recent literature makes between ordinary automation and autonomous experimentation.

If you're evaluating this as a company/investment/partnership landscape, I'd focus especially on Medra, Atinary, Potato, Chemspeed/SciY, and EOS, and then separately map the underlying robotics/instrument vendors. The interesting question is increasingly who owns the intelligence layer and the closed-loop data, rather than who owns the robot arm.

Yes. What you’re describing is essentially the “self-driving lab” / autonomous science stack:

scientific goal → model/agent proposes experiments → robot executes → instruments measure → data/model updates → next experiment That category is moving from academic demos toward real commercial infrastructure. A 2026 Nature Reviews Chemistry review describes the field as evolving toward multipurpose platforms where algorithms propose, execute, and interpret experiments with limited human intervention.

The companies I’d watch most closely

1. Atinary — probably the closest commercial analogue

Atinary is explicitly building the loop you described. Its Boston facility uses robotics and instrumentation to run Design–Make–Test–Analyze–Learn cycles, with experimental results automatically feeding ML models that select subsequent experiments.

This is particularly interesting if your target is chemistry, materials, formulations, catalysis, or pharma process development.

2. Medra — closest to an “AI scientist controlling a physical lab”

Medra is attacking the problem more from the agent/model side.

Its Physical AI Scientist combines an AI Experimentalist with physical laboratory automation. The company says its system can interpret literature and results, decide what to try next, execute experiments, perceive what is happening on the bench, and recover from errors.

If your mental model is:

“Give an AI scientist a research objective and let it figure out the experiments”

Medra is one of the most directly relevant companies.

3. Ginkgo — potentially the biggest-scale biology implementation

Ginkgo Bioworks is making a major push into autonomous labs in 2026. Its architecture combines Catalyst software with reconfigurable automation carts and dozens of instruments, with experimental data feeding AI/ML systems.

The interesting distinction is that Ginkgo isn't merely selling a robot. It's trying to turn the entire R&D operation into programmable infrastructure.

That's particularly relevant for:

• protein engineering
• cell engineering
• strain engineering
• biological screening
• high-throughput experimental biology

4. Arctoris — already very good at the model ↔ wet-lab loop

Arctoris operates an automated wet lab and explicitly supports integrating experimental results into active-learning pipelines. Its Ulysses platform provides machine-readable experimental data that can feed subsequent model iterations.

One particularly compelling example is its collaboration with IBM Research: computational molecular design proposes compounds, Arctoris synthesizes/tests them, and the resulting data feeds the next design cycle.

So Arctoris is closer to “autonomous drug-discovery factory” than generic lab automation.

5. Benchling — potentially the connective tissue

Benchling is coming at this from the software/data layer.

Its 2026 Automation product connects instruments, workcells, scientific records and analysis into a continuous loop. Benchling explicitly describes the architecture as AI/ML proposing candidates, automated labs executing them, results returning to the models, and the models selecting the next iteration.

This could become important because the hard problem isn't just the robot. It's:

model → protocol → orchestration → robot → instrument → data normalization → analysis → model

Benchling is trying to own that connective tissue.

6. Automata — the robotic infrastructure play

Automata is less of an AI scientist and more of the physical/software substrate on which one can run.

Its LINQ platform is designed around software-defined automation, interoperability and AI-driven workflows. It has also partnered with companies such as CellVoyant to connect predictive AI directly to robotic cell-culture execution in a closed loop.

Think:

AI brain: another company Automata: nervous system + muscles.

7. Plexymer — interesting vertically integrated newcomer

Plexymer explicitly combines experimental design, robotic execution, testing, structured data and ML into a single learning loop. Its focus spans biologic formulation and advanced materials.

It's worth watching because it's aiming at the full stack rather than selling just a robot or just an AI model.

8. CuspAI — interesting for materials

CuspAI is building an AI Materials Foundry connecting frontier models, computation, experimental laboratories and industrial partners. Its network is intended to close the loop from computational materials design through synthesis planning and experimental validation.

This is less "one giant autonomous robot lab" and more distributed AI → experiment infrastructure.

---

I would divide the market into four layers

The interesting thing is that no single company completely owns the stack yet.

🧠 1. Scientific reasoning / experiment selection

Companies developing the AI layer:

• Medra
• Atinary
• CuspAI
• various AI drug-discovery companies
• FutureHouse / Edison Scientific on the scientific-agent side

🦾 2. Physical execution

• Automata
• Ginkgo
• Arctoris
• Emerald Cloud Lab
• Opentrons
• Hamilton
• Chemspeed

🔄 3. Closed-loop orchestration/data

• Benchling
• Atinary
• Automata
• Ginkgo
• Emerald Cloud Lab

🧪 4. Vertical autonomous discovery

This is where things get especially interesting:

• drug discovery: Arctoris, Ginkgo, Medra
• chemistry/materials: Atinary, CuspAI, Plexymer
• cell biology: Medra, Ginkgo, Automata + CellVoyant
• protein/biological engineering: Ginkgo and emerging AI-bio companies

---

The real opportunity

If you're thinking about this as a company/market, I wouldn't frame it as "robotic labs."

The much more interesting product is:

AI Scientist → Experiment Compiler → Lab Operating System → Robot → Measurement → Learning

The robot is increasingly becoming a commodity-ish component. The scarce piece is the closed-loop intelligence and infrastructure that makes heterogeneous laboratory equipment behave like one programmable scientific computer.

That's also where the current research literature is pointing: the major bottlenecks are generalizability across workflows, interoperability between hardware, complete experimental provenance, and AI agents that can reason under uncertainty and safely control physical experiments.

And there is still a substantial gap between the vision and reality: even leading self-driving labs generally retain humans for oversight, exception handling, safety and scientific judgment.

If I were mapping the competitive landscape for an investment/company-building decision, my first five deep dives would be Medra, Atinary, Ginkgo, Arctoris, and Benchling. They represent five somewhat different bets on essentially the same emerging architecture.

This is less "one giant autonomous robot lab" and more distributed AI → experiment infrastructure.

---

I would divide the market into four layers

The interesting thing is that no single company completely owns the stack yet.

🧠 1. Scientific reasoning / experiment selection

Companies developing the AI layer:

• Medra
• Atinary
• CuspAI
• various AI drug-discovery companies
• FutureHouse / Edison Scientific on the scientific-agent side

🦾 2. Physical execution

• Automata
• Ginkgo
• Arctoris
• Emerald Cloud Lab
• Opentrons
• Hamilton
• Chemspeed

🔄 3. Closed-loop orchestration/data

• Benchling
• Atinary
• Automata
• Ginkgo
• Emerald Cloud Lab

🧪 4. Vertical autonomous discovery

This is where things get especially interesting:

The ecosystem bridging AI-driven experiment planning, robotic lab execution , and real-time model feedback (often termed Self-Driving Laboratories or Closed-Loop R&D ) spans cloud-based remote infrastructure, modular hardware automation, and AI-native discovery platforms.

The key players pioneering this space are categorized below by their operational approach:

1. Cloud Labs & Remote Robotic Execution Platforms

These companies provide fully digitized, remotely controllable wet labs where experimental protocols are translated directly from software/API commands into physical robotics.

• Emerald Cloud Lab : Operates a massive, standardized robotic facility that can be remotely programmed via code to execute end-to-end life science and chemistry protocols without human hands.
• Strateos : Offers smart, automated cloud laboratories combining robotic lab infrastructure with an integrated software platform for design, planning, and execution.
• Culture Biosciences : Focuses on biomanufacturing and fermentation, pairing automated bioreactors (Stratyx) with cloud software (Console AI) to enable remote, iterative bioprocess optimization.

2. AI-Native & Closed-Loop Discovery Platforms

These companies build the software and active-learning loops that actively generate hypotheses, design the next batch of experiments, and ingest the physical data back into the model.

• Kebotix : A pioneer in AI-driven self-driving labs for advanced materials and chemical discovery, coupling machine learning with automated robotic synthesis and characterization.
• Arctoris : Operates an automated platform for drug discovery combining robotics with real-time data capture to feed internal machine learning frameworks.
• Emerging AI/Lab Startups : Newer market entrants like Lila Sciences, Medra, and ReactWise are explicitly marketing closed-loop systems designed to auto-optimize synthesis parameters and shave massive timelines off experimental screening.

3. Hardware Automation & Workcell Integrators

These firms supply the physical arms, modular workcells, and foundational fluid-handling hardware that make automated orchestration possible.

• Opentrons : Provides flexible, open-access liquid-handling robots that lower the barrier to automating benchtop protocols and integrating with custom software workflows.
• Chemspeed : Builds high-throughput, automated robotic R&D manufacturing and dispensing workstations heavily utilized in automated chemistry and formulation.
• HighRes Biosolutions : (Frequently integrated into platforms like Benchling ) specializes in microfluidic and robotic workcell architecture designed for zero-click, high-throughput labs.

Are you looking to build a custom in-house automated workcell using modular hardware, or are you looking to outsource execution via a cloud lab API ? Let me know your target domain (chemistry, biologics, or materials science) to narrow down the ideal stack.

The convergence of AI model feedback, experimental planning, and robotic execution is the defining trend of Self-Driving Laboratories (SDLs) . Rather than treating automation as a static pipeline of scheduled liquid handlers, these systems use active learning, Bayesian optimization , or agentic AI to form a closed Design-Build-Test-Learn (DBTL) loop.

Companies operating in this space span specialized software orchestration, turnkey hardware-software integration, and AI-native therapeutics/materials discovery.

1. Dedicated Lab Orchestration & AI Platforms (The "Brain and Scheduler")

These companies provide the software and integration layers that translate high-level AI goals or model outputs into physical execution instructions across multi-vendor robotic workcells.

• Automata : Builds Software-defined LINQ , a process orchestration platform featuring LINQ Canvas (a node-based workflow builder) designed for parallel workflow execution, real-time analytics, and closed-loop error handling.
• Atinary Technologies : Focuses heavily on the intelligence layer for chemistry and materials science . Their SDLabs AI platform uses advanced machine learning and Bayesian optimization to design the next best experiment, connecting model optimization directly with automated screening and hardware execution.
• BioSero : Known for Green Button Go , an industry-standard software suite that schedules and orchestrates instruments from different robotic and analytical hardware vendors, allowing companies to build automated feedback-driven loops.
• LabVantage : Features LabVantage Cortex , which integrates native AI agents into LIMS and ELN workflows . It allows a scientist to state a goal, translates it into a multi-step execution plan, interfaces with physical robotic infrastructure, and feeds validated data back into the system's knowledge graph to refine future predictive accuracy.

2. Full-Stack / Turnkey Autonomous Labs (Hardware + AI)

These firms bridge the gap by deploying proprietary AI models alongside physical robotic workstations, automated guided vehicles (AGVs), and analytics suites.

• XtalPi : A prominent player in AI-driven drug discovery and materials science that offers an Autonomous Lab Solution . They link hundreds of robotic workstations, automated synthesis setups, and AGVs into a continuous, 24/7 closed-loop DMTA (Design-Synthesis-Test-Analyze) cycle.
• Chemspeed : Builds automated modular robotic workstations for parallel synthesis and characterization that are increasingly integrated with AI frameworks and active learning intelligence layers for autonomous execution.

3. Biopharma & Biotech Giants Building Closed-Loop Ecosystems

Many major life sciences organizations are transforming their internal R&D infrastructure into self-driving ecosystems or partnering heavily to achieve closed-loop execution.

• Recursion Pharmaceuticals : Operates massive wet-lab automation in tandem with high-throughput phenotypic screening . They have partnered with HighRes Biosolutions (and leverage platforms using orchestration via NVIDIA) to build self-driving labs that run millions of experiments weekly, directly updating predictive machine learning models.
• AstraZeneca : Developed its internal iLab automation program (in partnership with vendors like BioSero and Zinsser Analytic) to autonomously synthesize and purify small molecules in parallel based on real-time analytical feedback.

4. Legacy Automation Giants Adding AI Layers

Traditional lab automation hardware vendors are shifting past basic scheduling to support software-driven, feedback-based architectures.

• Thermo Fisher Scientific : Their Momentum and ARIA laboratory automation software platforms orchestrate complex multi-instrument workflows and integrate with expanding API ecosystems to handle closed-loop instructions.
• Hamilton Company & Tecan : Provide the baseline robotic liquid handling infrastructure that external AI orchestration platforms (like BioSero or custom python-driven APIs) hook into to achieve automated execution.

If you are looking to design or procure a system like this, could you share:

• What is your primary domain (small molecule chemistry, biologics/genomics, or materials science)?
• Are you looking for a turnkey platform or looking to integrate AI with existing hardware (like Hamilton/Tecan)?

The companies building this are generally grouped under self-driving labs (SDLs), autonomous science, AI experimentalists, or closed-loop R&D automation. The core idea is:

AI model → experiment design → robotic execution → measurement/data → model update → next experiment

The strongest players span several layers:

Full-stack autonomous labs (AI + planning + robots + feedback loops)

• Atinary Technologies — One of the clearest examples of a self-driving lab platform. It combines machine learning experiment optimization with robotics and lab instrumentation, targeting materials, chemistry, and pharma R&D.
• Lila Sciences — Building “AI Science Factories”: autonomous laboratories where AI systems generate hypotheses and run physical experiments across biology, chemistry, and materials science.
• Medra — Developing what it calls a “Physical AI Scientist”: systems that perceive lab environments, execute experiments, analyze results, and improve experimental design.
• Periodic Labs — Focused on combining AI researchers/agents with autonomous physical experimentation, especially in chemistry and physics domains.

AI experiment planning / scientific reasoning layer

• b12 Labs — Building AI agents that plan chemical synthesis and connect planning with robotic execution.
• Plexymer — Focuses on integrating experiment design, robotic build, testing, structured data, and machine learning into a continuous “design-build-test-learn” loop.

Robotic execution infrastructure

• Opentrons — A major provider of programmable lab robots and automation infrastructure. It is positioning its systems as the physical execution layer for AI-driven experimentation, including workflows where AI proposes experiments and robots run them.
• Chemspeed Technologies — Longtime automated chemistry platform provider moving toward AI-orchestrated self-driving laboratories.
• HighRes Biosolutions — Builds lab orchestration software and automation infrastructure connecting instruments, workflows, and AI-driven decision systems.

Cloud labs (remote robotic execution)

• Emerald Cloud Lab — Provides remotely operated automated laboratories where researchers submit experiments and receive data, an important foundation for scaling autonomous experimentation.
• Strateos — Similar direction: robotic labs accessible through software interfaces, enabling automated execution pipelines.

Adjacent enabling companies

• ABB — Supplies robotic systems used in some autonomous lab deployments, including collaborations with self-driving lab developers.
• Agilent Technologies and Mettler Toledo — Important instrument-layer players whose analytical hardware feeds data back into autonomous loops.

Research groups pushing the frontier

• Pacific Northwest National Laboratory — Developing autonomous science infrastructure and self-driving laboratory concepts at national-lab scale.
• OpenAI — Has explored AI-guided wet-lab automation, including systems combining language models, vision, and robotic execution for biological experiments.

The market map

A useful way to think about the stack:

The “holy grail” is not just automating pipetting or scheduling—it is a closed-loop scientific agent that can read literature, propose hypotheses, design experiments, run them overnight, interpret results, and decide what to try next. Several companies above are explicitly targeting that architecture, but the field is still early: reliability, data standards, instrument interoperability, and scientific reasoning remain major bottlenecks.

If you are looking from a venture/investment, partnership, or build-vs-buy perspective, the shortlist changes quite a bit.

Yes. What you’re describing is essentially a self-driving / autonomous laboratory: an AI system proposes experiments, robots execute them, the results are analyzed, and the resulting data updates the model/strategy for the next experiment. That closed loop is now becoming a real commercial category. www.nature.com

The companies I’d watch most closely

1. Medra — probably the closest match

Medra is unusually explicit about the architecture you're describing. Its AI Experimentalist turns research objectives into experimental plans; its Physical AI Lab executes them; results are interpreted and fed into subsequent experimental decisions. Its ML001 facility is running biology workflows including antibody discovery, protein engineering, gene editing and cell biology.

The interesting part is that they're not merely putting an LLM in front of a liquid handler. They're trying to make the AI reasoning layer and physical lab one system.

2. Lila Sciences — the most ambitious version

Lila is pursuing the broader "AI scientist" vision: models generate hypotheses, design experiments, run them in autonomous labs, observe the results and redesign experiments. Their AI Science Factory concept is explicitly intended to combine the computational and physical sides of science.

They're targeting life science, chemistry and materials, rather than just drug discovery.

3. Atinary — particularly interesting for optimization

Atinary is further along on the self-driving laboratory / Bayesian optimization side. Its Boston facility has autonomous platforms that continuously design, execute, analyze and learn from experiments, with integrations to robotics and instruments from companies such as ABB, Agilent, Bruker, Chemspeed and Mettler-Toledo.

I'd look at Atinary if your problem is something like:

"Given experimental results, what should the robot try next to optimize X?"

rather than primarily:

"Give an AI scientist general control over a biology lab."

4. Insilico Medicine — strongest vertically integrated pharma example

Insilico has gone unusually far in actually integrating AI drug-discovery systems with a physical automated lab. Its LabClaw system coordinates AI agents, LIMS, analysis pipelines and its Life Star 2 laboratory, creating a loop from computational predictions → experiments → analysis → updated biological priorities.

This is particularly compelling because the feedback loop isn't theoretical: it's embedded in an actual drug-discovery organization.

The infrastructure layer is also important

There are companies that aren't necessarily trying to be the "AI scientist" themselves, but are building the physical/software substrate that lets one exist.

Automata is one of the most interesting. Its LINQ platform connects instruments, robotic handling, workflows and structured data, explicitly positioning itself as an execution layer between AI and the physical laboratory. It is also working with NVIDIA on agentic laboratories.

Emerald Cloud Lab is the other one I'd pay attention to. ECL has >200 instrument types under a unified software interface, with essentially everything scriptable. That makes it an extremely useful physical execution backend for AI-driven experimentation.

And Benchling is attacking the data/control-plane problem. Its current stack connects experimental design, models, instruments, execution and results; its stated goal is specifically a loop where predictive models propose experiments, automation executes them, and results inform the next decision.

How I would map the emerging stack

Think of it as five layers:

Foundation models / scientific models
↓
Experiment planner / agent — Medra, Lila, Atinary, Insilico
↓
Lab orchestration — Automata, Benchling, ECL
↓
Robotics + instruments — Chemspeed, Hamilton, Tecan, Opentrons, ABB, etc.
↓
Experimental data → model feedback → next experiment

The really interesting companies are the ones trying to collapse those layers into one closed-loop system.

My shortlist if you're evaluating the space

I'd put Medra, Lila, Atinary, and Insilico in the "actually trying to build the autonomous scientist" bucket.

I'd put Automata, Emerald Cloud Lab, and Benchling in the "infrastructure that could become the operating system for autonomous science" bucket.

And there's an important distinction: most "AI + lab automation" companies still aren't truly autonomous. A genuine system has to decide what experiment to run next, not merely translate a human-written protocol into robot instructions. The literature is still describing fully autonomous experimentation as relatively narrow and domain-specific, which is why the companies above are interesting.

If you're thinking about this from an investment / competitive landscape perspective, I can also break these into who has the best technology, who has actual deployed autonomous labs, funding/valuation, customers, and where the white spaces are.

The ecosystem bridging AI-driven experimental planning, robotic execution, and real-time model feedback—often referred to as Self-Driving Laboratories (SDLs) or Closed-Loop Automation —is rapidly expanding across biotech, materials science, and chemistry.

Companies and institutions building these end-to-end loops fall into platform developers, life science pioneers, and industrial/materials innovators:

1. Life Sciences & Drug Discovery Platforms

• • Ginkgo Bioworks: Heavily invested in closed-loop automated infrastructure. They’ve partnered with major AI developers (like OpenAI) to link frontier models directly to their automated foundries for automated protein and cell engineering, significantly reducing production and screening costs.
• • Recursion Pharmaceuticals: Operates massive automated, high-throughput biology and chemical screening labs driven by deep learning. Their computer vision and neural networks analyze millions of cellular images per week, directly dictating the next iterative round of robotic assaying.
• • LabGenius: Operates EVA , a fully integrated digital-biological platform in London that utilizes machine learning and robotic automation to design, generate, and test complex therapeutic antibodies in a continuous learning loop.
• • XtalPi: Combines quantum mechanics, AI predictive modeling, and automated robotic workstations (often in partnership with hardware specialists like ABB) to rapidly synthesize, screen, and optimize drug candidates.

2. Chemistry, Flow Synthesis & Materials Science

• • Chemspeed Technologies: A premier hardware and orchestration provider that builds modular, high-throughput robotic research systems designed explicitly to integrate with AI-driven active learning and Bayesian optimization loops for organic and inorganic synthesis.
• • Periodic Labs: Focuses on automated materials-synthesis loops, pairing robotic furnaces, powder dispensers, and characterization tools with large language models and reasoning software to autonomously run hundreds of material variations daily.
• • Trilobio: Builds modular, biologically aware robotic hardware and software stacks designed to turn traditional static lab equipment into agile, self-driving biological experimentation units.

3. Enterprise & Institutional Orchestration

• • Sapio Sciences: Provides modern, AI-ready electronic lab notebooks (ELN) and laboratory information management systems (LIMS) designed to act as the "digital bridge" or data backbone that structures experimental feedback before passing instructions to hardware.
• • National Laboratories (e.g., Argonne National Laboratory, Lawrence Berkeley Lab): Public-private initiatives like the A-Lab or the Microbial Molecular Phenotyping Capability (M2PC) are pioneering autonomous mobile robotic scientists and open architectures for solid-state and energy storage discovery.

Are you looking to implement this loop for a specific domain—such as small-molecule chemistry, synthetic biology/protein engineering , or materials science —and do you need an evaluation of commercial hardware vs. cloud-lab APIs?

The landscape for automated lab planning and robotic execution tied to real-time model feedback—frequently called Self-Driving Laboratories (SDLs) or Closed-Loop Discovery Platforms —spans specialized AI software pioneers, hardware-software integration companies, and major life science/biotech operations.

1. AI-Driven SDL Software & Orchestration Platforms

These companies focus heavily on the "brain" and the digital middleware, using Bayesian optimization, active learning, or agentic AI to direct physical hardware.

• Atinary Technologies : A pioneer in self-driving labs and closed-loop AI optimization. Atinary provides its SDLabs agentic AI platform, which interfaces with hardware (such as Chemspeed and Bruker analytical tools) to autonomously run chemistry, materials, and formulation experiments, ingest negative data, and optimize parameters iteratively.
• Artificial : Specializes in whole-lab orchestration and scheduling software. They unify human personnel, robots, and complex analytic instruments, integrating tightly with AI models (like NVIDIA BioNeMo) to dynamically orchestrate multi-step drug discovery workflows.
• SciY / Chemspeed : Provides automated and robotic hardware synthesis solutions tied to advanced software platforms (SciY) designed to build high-throughput, closed-loop chemical and material development blocks.

2. Biology and High-Throughput Robotic Datacenter Builders

These organizations build physical closed-loop ecosystems at immense scales, combining proprietary wetware robotics with AI feedback loops.

• Vivodyne : Operates massive human biological datacenters that use robotics to run automated experiments on millions of living, vascularized human tissues. Their platform feeds causal, clinically concordant assay results directly back into computational models to make human biology computable.
• Ginkgo Bioworks : Extensively develops automated foundry and "cloud lab" capabilities. Ginkgo maps out structured autonomy frameworks for cell programming and synthetic biology, tying automated strain/enzyme screening results back into machine-learning generation loops.

3. Lab Automation Infrastructure & Hardware Integrators

These firms build the physical robotic tracks, modular cells, and device-agnostic middleware required to move plates, samples, and reagents without human intervention.

• Automata : Known for its LINQ lab automation architecture, Automata provides flexible, modular robotic automation lines designed to scale up throughput and easily integrate with third-party analytical and liquid-handling equipment.

4. Institutional & National Labs (Pioneering Reference Architectures)

While not commercial vendors, government and academic entities set the architectural standards for closed-loop cyber-physical systems:

• Argonne National Laboratory : Features massive autonomous discovery initiatives (such as Polybot and RAPID 200 ) integrating mobile robotic arms, glove boxes, and AI planning for energy storage and materials.
• Berkeley Lab : Home to the A-Lab , which uses autonomous robotic systems and automated powder handling/furnaces to synthesize and screen novel inorganic materials driven by live database feedback loops.

If you are designing your own stack, could you share whether you are focusing on chemistry/materials science or biological assays , and whether you need a vendor-agnostic orchestration software layer or turnkey physical lab hardware?

Yes. The category you’re describing is usually called self-driving labs (SDLs), autonomous experimentation, or increasingly Physical AI for science: a system that goes

model/hypothesis → experiment plan → robotic execution → measurement → interpretation → model update → next experiment.

As of 2026, there are several companies getting quite close to that full loop.

The companies I’d watch most closely

1. Medra — probably the closest match

Medra is unusually close to the architecture you described.

Its Physical AI Scientist explicitly combines an AI reasoning layer ("AI Experimentalist") with a physical robotic laboratory. The system is intended to design, execute, interpret, and improve experiments continuously, rather than merely generate protocols. Medra's ML001 lab is already running work in antibody discovery, protein engineering, gene editing and cell biology.

The interesting architecture is essentially:

scientific model → experimental design → robot → assay → result → reasoning model → revised experiment

That is much closer to an autonomous researcher than conventional lab automation.

2. Terray Therapeutics — an impressive vertical implementation

Terray Therapeutics is doing this specifically for small-molecule drug discovery.

Its EMMI platform combines generative chemistry models, predictive/selection models, automated experimentation, synthesis and assays. Critically, experimental data gets fed back into the models to drive the next design round.

Terray is also building TerraSynth, a synthesis-planning model that constrains AI-generated molecules by whether they can actually be made.

So rather than selling an autonomous lab as a generic platform, Terray is using the loop to run its own drug-discovery programs.

3. Atinary — the "self-driving lab platform" play

Atinary has been one of the earlier companies explicitly building the SDL category.

Its platform connects ML-based experimental design with laboratory robotics and instrumentation, allowing experiments to be designed, executed, analyzed and optimized in a closed loop. In 2026 it opened its own AI-powered Boston labs, integrating robots/instruments from companies including ABB, Agilent, Bruker, Chemspeed and Mettler-Toledo.

This is particularly interesting if you want infrastructure that other scientific teams can program, rather than a vertically integrated drug-discovery company.

4. Plexymer — very close to your exact formulation

Plexymer describes almost exactly the architecture in your question:

Design → Build → Test → Learn

It combines experimental design, robotic synthesis/formulation, characterization, structured data and ML, with the results used to select the next experiments. Its current focus is biologics formulation and advanced polymers/materials.

I'd put Plexymer high on the list if your interest is materials/formulation rather than drug discovery broadly.

5. Periodic Labs — the big "AI scientist" bet

Periodic Labs is particularly interesting because it's attacking the problem from the AI scientist side rather than simply making better lab automation.

Its thesis is that AI models need access to reality: autonomous labs generate experimental data, the AI proposes hypotheses, experiments test them, and the resulting data improves the AI. Periodic explicitly says it is building both AI scientists and the autonomous laboratories for them to operate in.

This is one of the companies I'd watch if you're interested in the eventual convergence of frontier models + robotics + scientific discovery.

6. Unchained Labs / Stuntman — execution infrastructure

Unchained Labs – Stuntman is approaching the problem from the automation side.

Stuntman embeds natural-language AI into a flexible laboratory automation system, including experiment planning and automated execution through Python/API control.

It's less "AI scientist discovers new science autonomously" and more AI-native programmable physical lab infrastructure, which could ultimately be an important layer of the stack.

---

There's an important distinction

I would divide the market into three layers:

A. AI scientist / reasoning layer

Companies like Periodic, 42, Discovery Loop, Medra

↓

B. Experiment orchestration / optimization layer

Companies like Atinary, Plexymer, Terray

↓

C. Physical execution layer

Companies like Medra, Stuntman, Emerald Cloud Lab, plus the underlying robot/instrument vendors.

The really valuable company may be the one that owns A+B+C simultaneously.

That's why Medra, Periodic and Terray stand out to me, albeit for very different reasons.

One especially important signal

This isn't just marketing anymore. The research literature is starting to demonstrate the same architecture: AI agents translating scientific goals into executable protocols, robotic execution, perception/quality control, and feedback-driven replanning. But there is still a substantial gap between a demo and a genuinely autonomous scientist. A 2026 stress test of LLM agents in a robotic chemistry lab, for example, found very low rates of expert-assessed executable workflows and essentially no successful workflow-level replanning from experimental feedback.

So the hard problem isn't really "can an LLM write a protocol?" It's:

Can an AI reliably maintain a scientific state, reason about uncertainty, choose the highest-value next experiment, translate it into physically valid robot actions, detect failures, interpret noisy measurements, and update its hypothesis without a human rescuing the loop?

That's the moat.

If you're looking at this from an investment / competitive landscape / "who should we partner with?" perspective, I'd narrow the field further to ~10 companies and map them across planning, foundation models, lab robotics, instruments, closed-loop optimization, biological vs. chemistry vs. materials, and actual autonomous execution.