SAION AI’s architecture follows the same pattern of cognition and execution, which it splits into three layers. Each layer corresponds to the different stages of a typical research workflow. The first layer analyses biological information, with the second step coordinating research tasks and attenuating software instances. The final stage puts laboratory work into physical actions.

The cognition layer of SAION uses data collected by Bota’s Cell2Cloud Biofoundry with extant scientific literature and publicly-available biological databases. The aim is to guide biological design decisions using patterns established by previous experiments and research. According to the company, its system can analyse various biological systems including genes and proteins.

Reviews of design-build-test systems suggest the combination of AI and robotic automation is one way that scientific research can be accelerated, although many implementations remain at the experimental stage.

An orchestration layer in SAION AI coordinates research tasks arbitrates between the software and hardware tools used by labs. Bota’s system uses a multi-agent framework of large language model reasoning to plan workflows and manage the interactions between specialised software tools. The platform includes more than three hundred scientific tools that cover tasks like sequence analysis and experimental design.

Systems combining multiple AI agents for research have begun to appear in various academic and experimental industrial projects. Platforms described in recent studies use coordinated agents to analyse literature, generate hypotheses, design experiments, and evaluate results, attempting to minimise the divide between computer modelling and laboratory execution.

The execution layer offered by the SAION AI connects experiments with laboratory hardware. It uses uses Bota’s Biological Protocol Language (one of several [PDF] such schemas) to convert plans into instructions for laboratory equipment. Data generated during experiments returns to the platform’s analysis layer, where it’s used to post-train the system’s models.

Closed loop experimentation has become central to laboratory automation research, with such lab systems maintaining the necessary continuity between models’ predictions and actual physical experimentation. In ideal scenarios, AI models propose experiments, robotic equipment perform them, and the results modify the next round of the AI’s hypotheses. Researchers note, however, that expert lab personnel need to supervise during the course of ongoing experiments.

Bota states that SAION AI has performed well on several life science benchmarks. According to the company, the platform achieved 70.7% on LitQA plus SuppQA tasks that measure comprehension of scientific literature. The system reached 88.2% accuracy on SeqQA benchmarks that test reasoning about DNA, RNA, and protein sequencing. It’s noted that large language models can be trained to perform exceptionally well on benchmark tests, however.

The company also states its systems complete research workflows comprising of analysis, experiment design, and execution with an accuracy of “above 90%.”

Several recent announcements illustrate similar attempts. In January 2026, Multiply Labs announced a collaboration with NVIDIA focusing on automated production of cell and gene therapies. The two companies in that instance plan to use AI, digital twins, and robotics to replace skilled manual labour.

Several research groups and industrial partnerships have begun work on developing AI-assisted bioreactor platforms that combine real-time sensing with machine learning models to, for example, adjust fermentation conditions during manufacturing.

According to the Wall Street Journal, Dash Bio has developed a system that automates the analysis of biological samples used in drug development similarly combining robotics, AI, and laboratory workflows to reduce the timescales of that aspect pharmaceutical research.

AI plays a growing role in wider drug design work. Alphabet subsidiary Isomorphic Labs has raised funds to develop systems to analyse biological data and propose candidate drug molecules. The company’s work will build on DeepMind’s AlphaFold protein structure prediction system, which reduced the time required to model protein structures.

Despite recent progress in computational design, no ‘AI-generated’ drugs have reached the market, with candidates remaining in clinical testing stages.

The global bio-manufacturing sector has expanded rapidly recently because the chemical and materials industries have adopted production methods common to biological research. Market estimates place the combined value of the sector above $1 billion for 2026, with continued growth expected as companies replace traditional chemical production research methods with those more associated with biological research methods.

SAION AI’s attempt to shorten the cycle between discovery and industrial production will, like similar projects, live or die depending on the complexity of the problem addressed. The often-high levels of complexity mean many experiments fail regardless of any involvement of AI or automation, although the industry seems firmly of the opinion that by converging AI models, laboratory robotics, and biological datasets, biotechnology R&D may be quicker for this generation than previously.

(Image source: “Biotechnology Program” by Hagerstown Community College is licensed under CC BY-NC-ND 2.0.)

Want to experience the full spectrum of enterprise technology innovation? Join TechEx in Amsterdam, California, and London. Covering AI, Big Data, Cyber Security, IoT, Digital Transformation, Intelligent Automation, Edge Computing, and Data Centres, TechEx brings together global leaders to share real-world use cases and in-depth insights. Click here for more information.

TechHQ is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.