The Wet Lab Gap: When AI Stops at the Bench Door
AI lab automation describes the digital and physical systems that allow algorithms and robotic instruments to design, run, and record experiments as one continuous, traceable workflow that turns computational ideas into tested results at scale. In life sciences, that promised bridge is still mostly missing. Coding agents write and test software, but AI co-scientists struggle once work leaves the screen and enters the wet lab. According to the Pistoia Alliance, only 1% of life science professionals report AI delivering meaningful value in wet lab settings, even though more than half see benefits around regulatory submissions and reporting. Most “AI scientist” offerings today remain trapped in simulation or analysis, unable to order reagents, run assays, or structure experimental data. The result is a sharp contrast between lively AI progress in software development and slow wet lab AI adoption where experiments still depend on manual planning and execution.
Why AI Co-Scientists Stumble on Physical Experiments
The core problem is translation: co-scientists can suggest hypotheses, but the physical lab work that follows is messy, fragmented, and often bespoke. Benchling co-founder Ashu Singhal points out that the hypothesis is often the easy part; the hard part is ordering materials, setting up notebook entries, running assays, capturing results, and feeding them back into the next experimental cycle. He describes experimentation as three buckets: repetitive, well-defined assays suited to workcells; experiments best sent to external CROs; and one-off, ad hoc work that will not pay back the cost of full automation. That last bucket is huge, which means human scientists will stay at the bench. Even where automated instruments or CROs exist, teams still struggle to push clean inputs to the hardware and turn raw outputs into structured data that AI systems can understand and reuse.
Lab Automation as the Missing Infrastructure for AI
For wet lab AI adoption to move beyond 1%, labs need infrastructure that connects models to instruments and data in both directions. Many AI co-scientists today are model-centric, with little visibility into what happens at the bench. Benchling is betting that life science automation needs three things: a tight connection to physical lab equipment, a structured data model rich enough for AI, and interfaces scientists accept as part of daily work. Its platform has evolved from digital record-keeping into a wiring layer for physical work: one-click ordering through partners like Twist Bioscience, Adaptyv, and Ginkgo Bioworks; a Model Hub that lets teams run structure prediction and generative models in context; and AI Connectors that use Model Context Protocol to link scientific records to external AI tools. Together, these pieces try to turn isolated robots and CROs into a coherent execution layer for AI-designed experiments.
Benchling Automation: Closing the Loop Between Code and Pipettes
Benchling’s new Automation product aims to close the loop between digital experiment design and physical execution. When a scientist or AI co-scientist designs an experiment, Benchling Automation sends the run to an appropriate workcell or instrument—through launch partners such as HighRes, Automata, Ginkgo Bioworks, Celltrio, Opentrons, and Hamilton—and then routes both raw and analyzed data back into the central notebook, tied to the samples that produced them. Prime Medicine’s work on PM359, a prime-edited therapy for chronic granulomatous disease, shows what this loop can look like in practice: Benchling’s AI Scientist synthesized data collected since 2022, mapped it to FDA validation requirements, designed targeted follow-up studies, and kept a living validation package, compressing months of work into days. Singhal describes this as a gated loop: AI drafts methods and entries, waits while humans run bench work, then resumes once new data arrive.
From Autonomous Labs to Human–AI Collaboration
Benchling’s strategy sits in a crowded field where startups and tech giants are racing toward more autonomous labs. Lila Sciences is building “AI Science Factories,” Medra is deploying robotic systems that can operate more than 75% of instruments scientists already use, Amazon has launched Bio Discovery, and companies such as Bristol Myers Squibb and Merck are rolling out enterprise AI tools and cloud commitments. In one project, Merck rebuilt bioanalytical testing for vaccine trials on Benchling, reporting a tenfold efficiency gain along with a 25% reduction in rote tasks and support for more than 1.1 million clinical samples. Singhal argues that high-quality, structured lab records are the fuel that makes AI co-scientists useful and that the destination is not a scientist-free lab, but an environment where AI and automation handle the drudge work while human judgment stays at the center.