Wet lab AI adoption: Powerful on paper, weak at the bench
AI lab automation is the use of artificial intelligence to design, schedule, and interpret experiments while directly coordinating physical laboratory instruments and workflows, turning digital suggestions into repeatable wet lab actions backed by structured data. For all the excitement around life science AI tools, most teams say they see little benefit at the bench. The Pistoia Alliance reports that only 1% of life science professionals currently see AI delivering measurable value in the wet lab, even though 54% report gains in regulatory submissions and reporting. That gap reflects a core problem: models are strong at analysis, weak at moving pipettes. Hypothesis generation, protocol drafting, and literature review sit comfortably in software. Ordering reagents, configuring assays, capturing instrument data, and tying results to exact samples remain stubbornly physical tasks that most AI systems do not touch.
The disconnect between AI theory and lab reality
Many laboratory AI integration efforts stall because they focus on models, not on the messy details of experiments. Tools like Google DeepMind’s Co-Scientist and OpenAI’s GPT-Rosalind show how AI can debate hypotheses or reason inside computational pipelines. But they rarely connect to the routine work of running assays day after day. As Benchling’s co-founder Ashu Singhal notes, designing a hypothesis is often the easy part; the hard part is ordering the right constructs, setting up notebook entries, executing the assay, and feeding data back into the next design loop. In that loop, every mislabelled tube or unstructured spreadsheet undermines trust in AI suggestions. Without a clean path from digital recommendation to physical execution and back, wet lab AI adoption lags and scientists default to manual methods they can audit by hand.
Lab automation as the bridge between AI and experiments
Lab automation offers a way to connect AI reasoning to the realities of sample plates and instruments. Singhal breaks work into three buckets: repetitive, well-defined assays suited to full automation; tasks that belong with external CROs; and one-off experiments that remain manual. Benchling Automation focuses on the first category by wiring instruments, workcells, and scientific records into one system. When a scientist designs an experiment, the platform can send runs to partners like HighRes, Automata, Opentrons, Hamilton, and others, then route both raw and analyzed data back to the exact notebook entry and samples. In this approach, AI does not float above the lab. It issues plans that are executable on real equipment, while automation handles formatting, scheduling, and data capture so that results are structured and traceable enough for the next AI-guided cycle.
Benchling’s AI Scientist: Grounding co-scientists in the physical world
Benchling’s AI Scientist architecture aims to embed an AI co-scientist inside day-to-day laboratory workflows. Model Hub lets teams run structure prediction and generative models on the same platform where they design constructs and track samples. AI Connectors, built on Model Context Protocol, tie scientific data to external AI tools. One-click ordering with Twist Bioscience, Adaptyv, and Ginkgo Bioworks reduces friction between design and materials. The AI Scientist then acts as a gated loop: it drafts methods and notebook entries, waits while humans or automated systems run the work, and resumes analysis once data flows back. Prime Medicine showed this in practice by using Benchling’s AI Scientist to map evidence for PM359 to FDA validation needs and maintain a living validation package. According to Benchling, this compressed a process that would have taken months into days.
A hybrid future: AI co-scientists, automated runs, and human judgment
The rise of AI lab automation is not about emptying benches but about changing how scientists spend their time. Singhal argues that many assays will remain one-off and ad hoc, where full automation makes little economic sense. Instead, platforms like Benchling focus on the repeatable core where AI and machines can excel. Early results hint at the impact: Merck’s regulated bioanalytics group rebuilt vaccine trial sample validation around automated Benchling workflows and reported a tenfold efficiency gain, along with a 25% reduction in rote tasks. As more instruments and CROs connect into unified platforms, life science AI tools can stop at recommendations less often and proceed to executable, auditable runs. That shift could turn AI from a distant theorist into a practical co-scientist that earns trust experiment by experiment.