When we published From Digital to Agentic: The 2026 Lab Operations Report in June, the central finding was a gap. More than 60% of the over 110 life sciences professionals we asked were exploring or piloting generative AI, yet only 5% had agentic AI in production. The barriers pointed less to a lack of ambition and more to disconnected systems, fragmented data and the governance needed to run AI at scale. That research captured a moment, and AI moves quickly enough that we wanted to know whether the picture had changed.
So, we asked again. At our 2026 Product Advisory Councils (PACs) in London, Basel, San Francisco and Boston we asked 89 customers* a series of live poll questions, which mirrored those in the report. The results are more nuanced than a simple story of acceleration, and in a few places, they support views we have held for some time, shaped by working alongside scientists and lab teams every day as a long-standing partner in the lab.
The clearest shift is in production use of generative AI (AI systems that generate, summarize, or interpret information in response to prompts or data input). In the report, 25% of respondents were using it in production. At the PACs the figure more than doubled to 52%. The share still exploring or piloting generative AI was 47%, compared with 60% in the report, which is what we would expect to see as pilots begin to graduate into production. Only 7% of attendees were not yet using AI at all.
In his blog From Experimentation to Agentic: What the Future Lab Looks Like, Chief Executive Officer Keith Hale wrote about the gap between intent and execution when it comes to AI. These numbers suggest that for generative AI at least, many of our customers are closing it. Moving from pilot to production is rarely a decision about the model itself. It usually reflects a decision that the data, the systems and the governance around it are ready to be trusted.
Agentic AI remains the smallest number in the data. The report found 5% using it in production and the PACs found 9%. Exploration is broader, with 32% exploring or piloting agentic AI against 27% in the report. A small number of organizations have started to deploy agents and a larger group is preparing to follow.
That pattern fits the comparison Keith draws with autonomous vehicles. A driverless car did not arrive overnight. It was built step by step, with safety systems layered on top of each other and trust earned before autonomy was extended. We expect the agentic lab to follow the same path. Our PACs showed us a handful of early adopters with the largest group close behind.
We asked where attendees currently use AI or plan to adopt it within the next 12 months. Data and analytics led at 73%, up from 57%, followed by workflow automation at 63% against 53%. The larger movements were elsewhere. Reporting and documentation rose from 27% to 54% and software development and scripting rose from 27% to 51%.
Reporting is a telling result. In the report, respondents also named it as the place where an AI agent would deliver the most value in their lab, at 58%. It is structured, repetitive and time-consuming work that takes scientists away from the science, and it appears to be where interest is turning into plans. The rise in software development and scripting is consistent with the growing use of AI coding tools to build the integrations that connected labs depend on.
System integration was a top investment priority for 50% of PAC attendees, in line with the same 50% in the report. Device automation (47%) and software automation (48%) were close behind, with AI-enabled software and embedded agents at 46%.
In Bridging the Gap: Why Integration and Automation are Holding Life Sciences Labs Back from AI we argued that connected systems and automated workflows are the structural prerequisites for agentic AI, and that automation can mean different things in different labs, from physical device automation to process and workflow automation. The near-equal split between device and software automation here shows both are being funded at the same time. Customers are not choosing between AI and the infrastructure underneath it. They are investing in both, because each depends on the other.
Having a data platform and using it well are still different things. Across the four PACs, 62% of attendees reported a data lake or analytics platform in place, with 44% describing it as established and actively used and 18% describing limited or inconsistent usage. Low quality or inconsistent data was cited as a barrier to AI by 26% of attendees, down from 42% in the report.
The gap between a platform that exists and one that people use is where connected data matters most. An agent can only act on what it can access and interpret, so a data platform that few people can access or query presents a ceiling on what AI can do.
In the report, governance and regulatory constraints ranked lower than the conversations we were having with customers would suggest, at 23%. At the PACs the figure was 33%. Privacy and security remained the most cited barrier at 52%, in line with the 58% in the report, lack of skills or training came in at 38%.
This fits our view that regulation and validation become more prominent as organizations move from pilots toward production. In a regulated lab, compliant systems, audit trails and clear human oversight are central to how work gets done, and the requirements only grow as AI takes on more autonomous tasks. It is a theme that comes up in many of our conversations with customers and is one we expect to rise further as agentic deployments increase.
Perhaps unsurprisingly for customers of lab software, attendees described themselves as further along than the report respondents. 59% were moderately digitalized, against 43% in the report, and 14% described their labs as fully digitalized, against 9%. Even so, we don’t see evidence of any organization running a truly autonomous lab.
The mood has shifted from whether to adopt AI toward how to run it in a way that is connected, governed and trusted. The central argument of the report still stands: the organizations that move fastest will be those that invest in connected data, integrated systems and clear governance alongside the AI itself.
It also reinforces that AI will support scientists and not replace them. The labs making the most progress are using automation and AI to remove the manual handoffs between machines, so that scientists can increase productivity and spend more of their time on hypothesis, insight and interpretation. We will continue to track how the picture develops.
*Results come from live polls of 89 Cenevo customers across the four PACs (all figures combine responses from the four events) held in September and October 2026, with more than 40% from large pharma organizations. This is a directional view of our customer base and not a measure of the whole industry.