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

The companies building this are usually described as self-driving labs (SDLs), autonomous labs, AI scientists, or closed-loop R&D platforms. The architecture you’re describing has four pieces:

1. AI experiment planner (LLM/agent + scientific models)
2. Experimental design optimizer (Bayesian optimization, active learning, simulation)
3. Robotic execution layer (liquid handlers, synthesis robots, instruments)
4. Feedback loop (results → model update → next experiment)

Key companies in this space:

Full-stack autonomous/self-driving labs

• Atinary Technologies — One of the clearest examples of the full closed-loop vision: AI proposes experiments, robots execute them, data feeds back into optimization models. Focus areas include chemistry, materials, and pharma R&D.
• Plexymer — Building a self-driving lab platform combining experiment design, robotic execution, testing, and machine learning feedback loops, with applications in materials and biology.
• Telescope Innovations — Developing AI-guided chemistry platforms and deploying self-driving laboratory systems for pharmaceutical and chemical research.
• XtalPi — AI + robotics company working on autonomous drug discovery and materials discovery workflows, including robotic experimentation.
• Edison Scientific / FutureHouse — Focused more on AI scientist agents and scientific reasoning systems, with efforts toward connecting AI research agents to experimental infrastructure.

AI-native biology labs

• Medra — Building what it calls a Physical AI Scientist: AI systems that interpret experiments, operate lab instruments, and continuously improve experimental design. Its focus is biology workflows such as antibody discovery and genomics.
• Arctoris — Automated drug discovery platform combining robotics, assays, and data generation for reproducible biology experiments.
• Strateos — Cloud laboratory infrastructure providing remote access to robotic biology labs and automated experimentation.

Robotics / execution layer companies

• Opentrons — Builds programmable laboratory robots and is positioning them as the physical execution layer for AI-driven science: AI agents generate plans, robots execute, and results return to models.
• Chemspeed Technologies — Robotic synthesis and automation platforms widely used in chemistry, materials, and pharma workflows.
• Tecan — Major laboratory automation company providing liquid handling and workflow automation hardware.
• ABB — Industrial robotics provider increasingly integrated into autonomous laboratory systems.

Large pharma / enterprise adopters

Several pharmaceutical companies are building internal autonomous R&D capabilities or partnering with SDL vendors. For example, Pfizer has deployed self-driving laboratory systems from Telescope Innovations.

Adjacent “AI scientist” infrastructure

These companies are attacking the reasoning/planning layer:

• Isomorphic Labs — AI-driven drug discovery focused on computational biology and molecular design.
• Recursion Pharmaceuticals — Combines biological experimentation, automation, and machine learning at scale.
• Insilico Medicine — Uses AI for target discovery and drug design.

How the market is splitting

A useful way to map the ecosystem:

The most complete realization of the “AI scientist → robot lab → model feedback” loop is currently being pursued by companies like Atinary, Medra, Plexymer, Telescope, Opentrons, and XtalPi—but the field is still fragmented, with different players owning different parts of the stack.

A related area worth watching is foundation models for science + physical AI, where the competition may shift from “automated protocols” toward systems that can autonomously generate hypotheses, design experiments, operate heterogeneous instruments, and learn from accumulated experimental data.

The convergence of automated lab planning, robotic execution, and closed-loop model feedback is often referred to as Self-Driving Laboratories (SDLs) or Cognitive R&D . Companies bridging the digital-physical gap use active learning or machine learning models to analyze experimental results in real time, plan the next iteration, and command robotic hardware to execute it without human intervention.

Key players leading this space across life sciences, chemistry, and materials science include:

• Atinary Technologies builds Atinary SDLabs and autonomous labs that couple proprietary machine-learning algorithms with automated synthesis and screening workflows. Their platform optimizes multi-objective R&D challenges in chemical, material, and life sciences by closing the loop between digital analysis and physical execution.
• Recursion Pharmaceuticals operates via the Recursion OS , a massive integrated techbio platform. They combine ultra-high-throughput automated wet-labs (capable of millions of cellular experiments per week) with massive supercomputing infrastructure to feed algorithmic models, identify biological relationships, and cycle back into automated physical testing.
• Emerald Cloud Lab provides a remotely accessible, fully automated Emerald Cloud Lab facility. While traditionally used for remote command-line execution of life science protocols, its programmatic and software-driven nature integrates natively with computational pipelines for automated, sequential experimental design.
• Strateos offers a Strateos Smart Lab Platform that unifies robotic cloud-based life science infrastructure with automated experimental design software. It allows researchers to programmatically run miniaturized, integrated biological and chemical workflows tied directly to data feedback loops.
• XtalPi deploys an XtalPi Autonomous Lab Solution that integrates predictive AI design software with physical automation units, mobile robots (AGVs), and robotic workstations. This creates a 24/7 closed-loop DMTA (Design-Synthesis-Test-Analyze) cycle primarily targeted at drug discovery and chemical development.

Are you looking to build an in-house closed-loop setup using commercial lab hardware (like Hudson, Tecan, or Hamilton arms), or are you looking for an external cloud-lab/SaaS partner to run these automated cycles for you?

Closed-loop, self-driving laboratories (SDLs) that combine AI-driven experiment planning, robotic execution, and active model feedback are built by a mix of specialized lab orchestration software providers, cloud biology infrastructure, and hardware robotics integrators.

The market ecosystem spans several key companies bridging the Design-Make-Test-Analyze (DMTA) loop:

Software Orchestration & Digital Integration Layers

• Benchling Automation : Connects electronic lab notebooks directly with instruments, synchronizing data models and queueing robotic execution orders across hardware partners.
• Anthropic Model Hardware Standard : Provides standard interfaces allowing AI agents to control multi-vendor hardware (liquid handlers, microscopes, and robotic arms) natively.

Robotic Execution & Lab Infrastructure Providers

• Automata : Builds modular, flexible lab automation infrastructure and robotic workcells tailored for high-throughput, closed-loop iterations.
• HighRes Biosolutions : Designs integrated robotic screening systems and dynamic scheduling software (Cellario) used to orchestrate complex hardware paths.
• Hamilton Robotics & Mettler Toledo / ABB : Provide core robotic liquid handling, automated weighing, and physical synthesis systems engineered for continuous, lights-out 24/7 execution.

Cloud Labs & End-to-End Autonomous Facilities

• Ginkgo Bioworks : Integrates massive horizontal foundry automation with software pipelines to run automated organism engineering and design loops.
• Dunia : Developing massive automated infrastructure like the planned Berlin GigaLab to execute closed-loop solid-state and materials discovery.

Are you looking to integrate this for biological discovery/liquid handling, chemical synthesis , or materials science ? I can dive deeper into the specific hardware-software stack best suited for your domain.

The field you’re describing is usually called self-driving labs (SDLs), autonomous labs, or closed-loop AI experimentation: systems where an AI model proposes experiments, robots execute them, instruments generate data, and the results update the model for the next experiment. The core loop is often called Design → Make → Test → Analyze → Learn (DMTA-L).

Companies building pieces of this stack include:

Full-stack autonomous science / “AI scientist + robot lab”

• Atinary Technologies — Builds AI-driven self-driving labs with machine learning optimization, robotics integration, and closed-loop experimental control. Their SDLabs platform is aimed at chemistry, materials, and R&D workflows.
• Medra — Building a “Physical AI Scientist” platform that combines scientific reasoning models with autonomous lab execution. Their system targets biology workflows such as antibody discovery, protein engineering, and cell biology.
• Plexymer — Focuses on autonomous R&D loops combining experimental design, robotic execution, testing, structured data, and machine learning, especially for materials and biologics applications.
• Discovery Loop — A newer AI-for-science company aiming at large-scale automated discovery and parallel experimentation.

AI experiment planning / scientific agents

• b12 Labs — Building AI agents that can plan chemistry experiments and connect those plans to robotic execution, targeting pharma and chemical synthesis workflows.
• Scivity — Building agent-based research automation: planning, execution, validation, and reporting, initially focused on computational workflows with physical integrations developing.

Robotics + lab execution infrastructure

• Opentrons Labworks — Provides programmable liquid-handling robots and AI-oriented lab automation infrastructure. It is a key hardware layer for many autonomous-lab approaches.
• Chemspeed Technologies — Builds modular robotic chemistry automation systems used in self-driving laboratory setups.
• HighRes Biosolutions — Provides orchestration software connecting instruments, robots, scheduling, and workflows.

Data / orchestration layer

• SciY — Focuses on lab digitization, scientific data infrastructure, and AI-ready laboratory operations, including self-driving lab architectures with Chemspeed.
• Bruker — Through Chemspeed and SciY initiatives, is moving toward integrated automation + analytics + AI orchestration for R&D.

Adjacent / enabling companies

• Benchling — Builds research data management infrastructure (ELN/R&D systems) that can become part of automated biology workflows.
• Labguru — Provides digital lab management infrastructure.
• Agilent Technologies, Thermo Fisher Scientific, and Danaher Corporation — Major instrument and automation vendors whose hardware is increasingly being integrated into autonomous workflows.

How the stack is shaping up

A typical autonomous lab stack looks like:

1. Foundation/scientific model

• Generates hypotheses, reads literature, proposes experiments.

2. Experiment planner

• Converts goals into protocols, constraints, and optimization strategies.

3. Lab orchestration layer

• Schedules robots, instruments, samples, and workflows.

4. Robotic execution

• Liquid handlers, robotic arms, synthesis systems, incubators, microscopes.

5. Measurement + data layer

• Instruments generate structured data.

6. Learning loop

• Models update and select the next experiment.

The biggest strategic question is whether the winner is a “robotics company with AI” (hardware-first), an “AI scientist with access to labs” (model-first), or a platform layer that orchestrates everyone else’s hardware. Right now the market has strong contenders in each category.

If you’re looking specifically at drug discovery, materials discovery, synthetic chemistry, or biotech wet labs, the company landscape differs quite a bit.

Companies building closed-loop, automated "self-driving laboratories" (SDLs)—combining AI planning, robotic execution, and iterative model feedback—span modular lab orchestration software, hardware integration, and full-stack autonomous R&D .

Key Companies and Platforms

If you would like to narrow this down, let me know:

• Are you focusing on biology/genomics or chemistry/materials science?
• Are you looking to build an in-house automated cell or use a remote cloud lab?

I can provide a tailored architectural blueprint or specific integration steps.

Yes. This is becoming a distinct category: autonomous / self-driving labs or lab-in-the-loop scientific discovery. The important distinction is whether a company merely automates execution, versus actually closing the loop:

model proposes experiment → robot executes → instruments generate data → model updates → next experiment is selected.

The companies I'd watch most closely are:

The most interesting ones

Lila Sciences is probably the one I'd investigate first if you're thinking about a company whose fundamental product is AI + autonomous laboratory. NVIDIA explicitly describes Lila as scaling the scientific method with automated labs that generate data and validate experiments designed by its AI system.

LabGenius is a particularly clean proof point. Its EVA platform uses ML to design antibody candidates, automated screening to test them, and the resulting experimental data to update its optimization process. The company describes this explicitly as a closed-loop discovery platform.

Emerald Cloud Lab is different but extremely important. It has >200 instrument models under a unified software interface and can execute experiments remotely, with experiment data automatically organized into its knowledge graph. It's essentially infrastructure for giving an AI system a very large physical laboratory to control.

Opentrons is attacking the robotics side. In 2026 it added simulation/visualization for AI-generated workflows, partnered with NVIDIA on physical-AI training, and partnered with HighRes on an AI-agent-to-agent laboratory workflow.

And Benchling is moving into the orchestration/data side: its 2026 Automation product explicitly targets the "lab-in-the-loop" architecture where AI/ML proposes experiments, automated labs execute them, and the resulting data goes back into the models.

I would actually map the market as a stack

Your desired system looks roughly like:

                 SCIENTIFIC MODEL / AGENT
                         │
                 hypothesis / design
                         │
                         ▼
              EXPERIMENT PLANNER
                         │
             protocol + parameters
                         │
                         ▼
        ┌────────────────────────────┐
        │   LAB ORCHESTRATION LAYER  │
        │ Benchling / HighRes / etc. │
        └──────────────┬─────────────┘
                       │
                       ▼
             ROBOTIC EXECUTION
          Opentrons / ECL / HighRes
                       │
                       ▼
             INSTRUMENTS + ASSAYS
                       │
                       ▼
                 EXPERIMENTAL DATA
                       │
                       ▼
             ANALYSIS / KNOWLEDGE
                       │
                       └──────────────► MODEL
                                         │
                                         ▼
                                  NEXT EXPERIMENT

The really valuable company, in my view, isn't necessarily the robot manufacturer. It's the system that owns the decision-making loop and can dynamically decide what experiment to run next based on what just happened.

That's where the distinction between "automated lab" and autonomous scientist emerges.

There's also significant movement toward making the hardware itself agent-accessible. For example, Anthropic announced a 2026 Model Hardware Standard intended to let AI agents operate multiple laboratory instruments and robotic devices, reason through experimental steps, adjust parameters in real time, and sometimes recover from hardware errors.

If you're evaluating this as a startup/investment opportunity, I'd separate the landscape into (1) autonomous discovery companies, (2) cloud labs, (3) robotics, (4) lab orchestration, and (5) scientific foundation models. The competitive gaps between those layers are actually quite interesting.

The shift toward closed-loop, self-driving laboratories (SDLs) —where AI plans experiments, robotics execute them physically, and real-time data feeds back into the models to update the next hypothesis cycle—is rapidly moving from research concepts into commercial deployment.

The ecosystem of companies and platforms building this stack spans orchestration software, hardware robotics integration, and AI-native therapeutics or materials discovery.

1. Lab Orchestration & Scheduling Software (The "Brain" & Traffic Control)

These platforms sit between the AI model and the physical hardware, translating digital experimental designs into robotic movements and managing dynamic scheduling.

• Biosero: Maker of Green Button Go (GBG) software. Biosero specializes in device-agnostic orchestration, integrating AI assistants to dynamically plan, schedule, and optimize multi-instrument workcells (liquid handlers, robotic arms, readers) in real time.
• HighRes Biosolutions: Known for modular, micro-solid-phase and liquid-handling robotic systems (like the Cellario software ecosystem and Access-laboratory robots) heavily utilized in high-throughput pharma automation.
• Retisoft: Provides Genera orchestration software, emphasizing advanced look-ahead workflow analysis and deadlock prevention for complex, automated instrument chains.

2. AI-Native & Fully Automated Biotechs (Closed-Loop Discovery)

These organizations have built massive in-house physical infrastructure tightly coupled with machine learning loops to run thousands to millions of experiments autonomously.

• Recursion Pharmaceuticals: Operates massive, high-throughput automated biology labs running millions of wet-lab experiments weekly. They partner with companies like NVIDIA and HighRes Biosolutions specifically to construct scalable, self-driving, AI-orchestrated infrastructure.
• Atinary: Focuses on autonomous chemistry and materials synthesis. Their closed-loop AI platform integrates robotic synthesis cells with analytics to autonomously run experiments, process failure/negative data, and iterate parameters without human bottlenecks.
• IBM Research (Robo RXN): A pioneer in applying natural language processing (treating chemical formulas as text/sentences) to map reactions directly to robotic execution hardware for autonomous chemical synthesis.

3. Big Pharma In-House Closed-Loop Initiatives

Major pharmaceutical companies are building proprietary closed-loop infrastructure by partnering with specialized automation vendors:

• AstraZeneca: Operates its third-generation iLab automation platform (built alongside vendors like BioSero and Zinsser Analytic) to synthesize and purify small-molecule compounds in parallel loops.
• Eli Lilly & Merck (MSD): Heavily investing in physical automation paired with agentic AI platforms (such as Merck's large-scale deployments with Google Cloud) to digitize and automate the Design-Build-Test-Learn (DBTL) cycle.

Are you looking to implement this stack for biology (e.g., cell painting, high-throughput screening) or chemistry/materials science (e.g., synthesis, formulation optimization) ? I can share more details on how the software layer integrates with specific hardware vendors.

Yes. The category you’re describing is generally called self-driving labs / autonomous science / physical AI for science: an AI system proposes experiments, software translates them into executable protocols, robots run them, instruments generate data, and the results feed back into the model to choose the next experiment.

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

The companies I’d look at first

1. Atinary — probably the cleanest match

atinary.com is explicitly building Self-Driving Labs. Its system combines ML-based experiment planning with robotics and instrumentation, and its Boston facility runs closed-loop Design → Make → Test → Analyze → Learn cycles. In 2026 it opened a physical AI-powered laboratory integrating equipment from ABB, Agilent, Bruker, Chemspeed and Mettler-Toledo.

This is essentially:

scientific objective → model chooses experiment → robot executes → instruments measure → model updates → next experiment

That's almost exactly your description.

2. Lila Sciences — the "AI scientist + factory" vision

lila.ai is pursuing a more vertically integrated version: AI models coupled to robotic laboratories that can conduct experiments continuously across biology, chemistry and materials science. The company's "AI Science Factories" are intended to generate proprietary experimental data while the AI controls increasingly large portions of the discovery loop.

I'd put Lila in the most ambitious "build the entire autonomous scientist" bucket.

3. Medra — particularly interesting for agentic lab control

medra.ai is building a Physical AI Scientist rather than merely an optimization algorithm. Its 2026 system includes an "AI Experimentalist" designed to translate natural-language scientific objectives and human protocols into executable experiments, measure the results, learn from them and improve subsequent experiments.

That makes Medra particularly interesting if by "model feedback" you mean an AI agent actually deciding what the robot should do next, rather than Bayesian optimization sitting on top of a fixed workflow.

4. Emerald Cloud Lab — probably the strongest execution substrate

emeraldcloudlab.com is a slightly different beast. ECL has built a gigantic software-controlled laboratory with 200+ instrument types and a unified API/software environment. Scientists can remotely specify experiments and have robots execute them.

What's especially relevant is that ECL explicitly describes LLM-assisted experiment design and closed-loop optimization as an AI direction, including an API for AI systems to both read experimental data and write commands to execute new experiments.

So I'd think of ECL as:

"the cloud compute layer for physical experiments."

That is potentially extremely valuable if your goal is to let external AI models control experiments.

5. Telescope Innovations — chemistry-focused full loop

telescopeinnovations.com is building self-driving chemistry laboratories with AI-guided experimentation, process analytical technology and robotic execution. It installed a second self-driving lab at Pfizer in 2026.

This is one I'd investigate if your application is synthetic chemistry / process chemistry / pharmaceutical development rather than general biology.

6. Radical AI — materials

radical.ai is pursuing essentially the same architecture for materials. Its AI agent ("Antimatter") proposes materials, robots synthesize them, testing generates experimental data, and the results go back into the agent for the next iteration.

That's a particularly clean example of the model → physical experiment → feedback → model loop.

7. Plexymer

plexymer.com describes its platform almost exactly in these terms: intelligent orchestration, robotic experimentation, application-specific testing, structured data and machine learning operating as a single learning system. Its focus includes biologic formulation and advanced materials.

---

There's another important layer: the "robot operating system"

Companies don't necessarily need to build the entire autonomous scientist themselves.

Opentrons is especially interesting here. opentrons.com has a large installed base of standardized lab robots and is explicitly working with NVIDIA on AI-enabled laboratory robotics and physical-AI training data. Its vision is essentially to provide standardized physical infrastructure through which AI-designed experiments can be executed reproducibly.

And Chemspeed is moving toward an open self-driving-lab platform combining automation, analytics, lab management and AI orchestration.

This suggests a stack emerging roughly like:

                 AI Scientist / Agent
                         │
            literature + scientific reasoning
                         │
                  experiment planner
                         │
              Bayesian/ML optimization
                         │
              Lab orchestration layer
                         │
        ┌────────────────┼────────────────┐
        ↓                ↓                ↓
   liquid handler    synthesis robot    analytical
      / robot            / reactor       instruments
        │                │                │
        └────────────────┼────────────────┘
                         ↓
                 experimental data
                         │
                  analysis / QC
                         │
                         └──────→ AI Scientist
                                  ↑
                                  └── next experiment

The really interesting companies

If I were mapping the competitive landscape specifically around "AI plans experiments → robot executes → data feeds model → AI decides next experiment", my shortlist would be:

Tier 1 — closest to the complete vision

• Atinary
• Lila Sciences
• Medra
• Telescope Innovations
• Radical AI

Tier 2 — critical infrastructure / platforms

• Emerald Cloud Lab
• Opentrons
• Strateos
• Chemspeed

Tier 3 — domain-specific autonomous discovery

• Plexymer — biologics/materials
• Kebotix — materials
• XtalPi — computational + automated drug discovery
• LabGenius — AI/robotic protein engineering

There is also a substantial academic ecosystem around this. The field has progressed from narrow automated workflows toward multipurpose systems in which algorithms can propose, execute and interpret experiments with limited human intervention.

One distinction I'd make

There are really three different businesses hiding under "automated lab":

1. AI experimentalist — decides what experiment should happen next.
2. Lab OS/orchestration — converts that decision into executable instructions across heterogeneous instruments.
3. Robotic lab/cloud lab — physically performs the experiment and returns structured data.

The companies that can own all three layers have the most interesting strategic position. Atinary, Lila, Medra and some of the newer autonomous-lab companies are explicitly trying to get there, while ECL, Opentrons, Chemspeed and Strateos are particularly important pieces of the infrastructure.

If you're evaluating this as a market/company-building opportunity, I can also map ~20 companies by stack layer, modality (biology/chemistry/materials), funding, customers, and how autonomous their closed loop actually is.

I'd put Lila in the most ambitious "build the entire autonomous scientist" bucket.

3. Medra — particularly interesting for agentic lab control

medra.ai is building a Physical AI Scientist rather than merely an optimization algorithm. Its 2026 system includes an "AI Experimentalist" designed to translate natural-language scientific objectives and human protocols into executable experiments, measure the results, learn from them and improve subsequent experiments.

That makes Medra particularly interesting if by "model feedback" you mean an AI agent actually deciding what the robot should do next, rather than Bayesian optimization sitting on top of a fixed workflow.

4. Emerald Cloud Lab — probably the strongest execution substrate

emeraldcloudlab.com is a slightly different beast. ECL has built a gigantic software-controlled laboratory with 200+ instrument types and a unified API/software environment. Scientists can remotely specify experiments and have robots execute them.

What's especially relevant is that ECL explicitly describes LLM-assisted experiment design and closed-loop optimization as an AI direction, including an API for AI systems to both read experimental data and write commands to execute new experiments.

So I'd think of ECL as:

"the cloud compute layer for physical experiments."

That is potentially extremely valuable if your goal is to let external AI models control experiments.

5. Telescope Innovations — chemistry-focused full loop

telescopeinnovations.com is building self-driving chemistry laboratories with AI-guided experimentation, process analytical technology and robotic execution. It installed a second self-driving lab at Pfizer in 2026.

This is one I'd investigate if your application is synthetic chemistry / process chemistry / pharmaceutical development rather than general biology.

6. Radical AI — materials

radical.ai is pursuing essentially the same architecture for materials. Its AI agent ("Antimatter") proposes materials, robots synthesize them, testing generates experimental data, and the results go back into the agent for the next iteration.

That's a particularly clean example of the model → physical experiment → feedback → model loop.

7. Plexymer

plexymer.com describes its platform almost exactly in these terms: intelligent orchestration, robotic experimentation, application-specific testing, structured data and machine learning operating as a single learning system. Its focus includes biologic formulation and advanced materials.

---

There's another important layer: the "robot operating system"

Companies don't necessarily need to build the entire autonomous scientist themselves.

Opentrons is especially interesting here. opentrons.com has a large installed base of standardized lab robots and is explicitly working with NVIDIA on AI-enabled laboratory robotics and physical-AI training data. Its vision is essentially to provide standardized physical infrastructure through which AI-designed experiments can be executed reproducibly.

And Chemspeed is moving toward an open self-driving-lab platform combining automation, analytics, lab management and AI orchestration.

This suggests a stack emerging roughly like:

                 AI Scientist / Agent
                         │
            literature + scientific reasoning
                         │
                  experiment planner
                         │
              Bayesian/ML optimization
                         │
              Lab orchestration layer
                         │
        ┌────────────────┼────────────────┐
        ↓                ↓                ↓
   liquid handler    synthesis robot    analytical
      / robot            / reactor       instruments
        │                │                │
        └────────────────┼────────────────┘
                         ↓
                 experimental data
                         │
                  analysis / QC
                         │
                         └──────→ AI Scientist
                                  ↑
                                  └── next experiment

The really interesting companies

Yes. What you’re describing is usually called a self-driving lab, autonomous lab, or closed-loop/lab-in-the-loop experimentation:

Model proposes experiment → planner chooses next experiment → robots execute → instruments generate data → model updates → next experiment. As of 2026, there are several companies attacking different parts of this stack.

The companies I’d look at first

The most directly relevant

Atinary is probably the closest conceptual match to what you wrote. Its platform is explicitly designed around ML selecting experiments, robotics performing them, and experimental outcomes feeding back into the model. It has also moved from software into physical “Scientific Discovery Factories.”

Medra is another particularly interesting one. Its “Physical AI Scientist” combines an AI scientific-reasoning layer with its own physical lab. The stated loop is essentially design → execute → interpret → improve, including assay development and parameter tuning rather than merely executing predefined protocols.

Plexymer is similarly aiming at the full loop: experimental design, robotic build, analytical testing and ML-driven learning, with applications spanning biologic formulation and advanced materials.

The infrastructure play

If you aren't trying to buy an entire autonomous laboratory and instead want to build the system yourself, I'd pay particular attention to Opentrons + Automata + Benchling.

• Opentrons is increasingly positioning its robots as the physical execution layer for AI. Its current stack includes OpentronsAI and an MCP server, and its partnership with HighRes demonstrates agents translating natural-language intent into executable laboratory workflows.
• Automata is building something closer to a lab operating system: robotics, scheduling/orchestration, workflow execution and data infrastructure, with MCP connectivity specifically intended to connect AI models.
• Benchling has become substantially more relevant here in 2026. Its new Automation product connects instruments/workcells with scientific records, automatically returns experimental results, and is explicitly designed to close the lab-in-the-loop. Its launch ecosystem includes Automata, Ginkgo, Opentrons and Hamilton.

Synthace is worth considering if your bottleneck is experimental design and translating it into reliable instrument instructions. It handles multivariate/DOE planning, generates automation instructions, and automatically structures the resulting data.

And Emerald Cloud Lab is an interesting option if you want to outsource the physical infrastructure entirely: experiments are scriptable and executed remotely across a broad set of automated techniques.

One important distinction

There are really three markets converging here:

1. AI scientist / experiment planner Atinary, Medra, Plexymer
2. Lab OS / orchestration layer Automata, Benchling, Synthace, HighRes
3. Physical execution / cloud lab Opentrons, Emerald Cloud Lab, Arctoris, Ginkgo and traditional automation vendors

The interesting opportunity is the connective tissue between all three.

A truly autonomous system needs something like:

Frontier model / domain model ↓ Experiment planner + hypothesis generator ↓ Constraint/safety/reagent-aware compiler ↓ Lab orchestration layer ↓ Robots + instruments ↓ Structured experimental data ↓ Analysis / Bayesian optimization / learned scientific model ↺ planner

That's a materially different product from “AI that writes a pipetting protocol.”

And the industry is moving in exactly that direction: for example, Anthropic just announced a Model Hardware Standard intended to let AI agents interact with programmable scientific hardware such as microscopes and robotic arms.

If you're evaluating this as a build-vs-buy decision, I would shortlist Atinary, Medra, Automata, Opentrons and Benchling first. The right choice depends heavily on whether your domain is drug discovery, protein engineering, cell biology, chemistry/materials, or something else.

The convergence of AI planning, robotic hardware, and real-time model feedback is driving the rise of Autonomous / Self-Driving Laboratories (SDLs) . Rather than just executing fixed protocols faster, these systems close the loop: an AI model proposes a hypothesis or optimization path, robotics execute the physical experiment, data is parsed instantly, and the model updates itself for the next cycle.

Companies and entities building this closed-loop stack span modular hardware providers, cloud-based infrastructure, and stealth-to-market AI-physical science startups:

• Chemspeed Technologies: A premier provider of modular, high-throughput robotic R&D and QC systems that integrate with predictive software to function as vendor-agnostic self-driving labs for chemistry and materials science.
• Opentrons Labworks: Known for accessible benchtop liquid-handling robots (like the OT-2 and Flex), Opentrons provides the open-source software flexibility heavily leveraged by labs building custom closed-loop automated workflows.
• Lila Sciences (Flagship Pioneering): Launched out of Flagship's incubator model, Lila Sciences focuses explicitly on building autonomous labs spanning chemical, material, and life sciences by tightly coupling automated execution with generative AI.
• Medra: Emerging in the "Physical AI Scientist" space, Medra builds integrated robotic systems designed specifically to automate and close the loop for biopharma R&D workflows.
• IBM (Robo RXN): While a tech giant rather than a pure-play lab automation startup, IBM’s Robo RXN platform maps chemistry via natural language processing, translating AI-predicted reaction outcomes directly into execution code for robotic synthesis hardware.
• Academic & Institutional Biofoundries (e.g., University of Illinois iBioFAB): Though academic, platforms like iBioFAB pair automated robotic liquid handlers with Bayesian optimization algorithms (like BioAutomata) to run complete design-build-test-learn cycles autonomously, setting the architectural blueprint that commercial entities emulate.

If you'd like to narrow this down, tell me:

• Are you looking for chemistry/materials synthesis or biology/high-throughput screening?
• Do you need benchtop modular units for an in-house lab or a fully remote cloud lab infrastructure?

I can tailor the vendor list and architectural recommendations accordingly.