In his remarks at SapioCon 2026, Kevin Cramer, founder and CEO of Sapio Sciences, put a useful frame around where our industry may be headed.

“The marriage of in-silico, smart scientists, and wet lab work is the sweet spot for accelerating the path from hypothesis to evidence to next-step decisions.”

The line stuck because it captures the central tension being experienced by labs and scientists today. While AI is everywhere and investment is rising, the daily experience in the lab often remains stubbornly manual.

And so, SapioCon 2026 set out to investigate not whether advanced AI methods exist, but whether they can be made operational inside workflows that are governed, traceable, and built for real-world decisions.

The coordination constraint

Becky Upton, president and CEO of the Pistoia Alliance, opened the event by naming a reality the industry sometimes skirts. Ambition has rarely been the bottleneck in R&D. Coordination is.

Upton cited a familiar industry estimate that as much as 80 percent of a data scientist’s time is spent collecting and wrangling data rather than generating insights. Mike Hampton, Sapio’s Chief Commercial Officer, translated this into a lab reality of “trapped science,” meaning digitization that records a narrative but fails to keep work moving beyond silos.

That gap helps explain the rise of shadow AI. Scientists are already reaching for external tools outside governed environments because they need speed and interpretation support closer to the experimental record.

In Sapio’s recent survey, more than three-quarters of scientists admitted to using these unvetted tools just to get their work done. The takeaway was clear: we don’t need more isolated tools; we need workflows that bring intelligence into the governed record where it can be monitored, reused, and audited.

Operational evidence at scale

The most convincing moments were the ones that showed intelligence embedded in the record rather than bolted on as an afterthought.

Rob Brown, head of the Scientific Office at Sapio Sciences, demonstrated this in Sapio ELaiN using a 28-page SOP for digital droplet PCR. Traditionally, turning a document like that into a usable, multi-tab experiment template takes hours of configuration. In the session, ELaiN generated a structured template directly from that SOP in roughly seven minutes, complete with protocol overviews, plate layout logic, and instrument run settings.

The power of this digital-first approach was echoed in our customer success stories. Federico Lionetti from Navignostics described a spatial single-cell proteomics platform where treatment outcomes were three times more effective than standard care in a 120-patient study. This was followed by Amy Yeung from the Wellcome Sanger Institute sharing the scale of an operation that sequences the equivalent of a gold-standard human genome every 12 minutes, emphasizing that operational data must become a strategic asset rather than just a log of past events.

Designing AI to survive the real world

Andreas Steinbacher of Novo Nordisk voiced an uncomfortable truth: AI cannot fix what humans have not agreed on. He argued for a preclinical FAIR data layer, one that separates data capture from data management to support scale across global research sites. This is connected to a broader theme shared by our partners: the data layer is the AI strategy. 

Christine Tsien Silvers from AWS made the case for verification approaches that ensure results are factual rather than hallucinations, which is critical in regulated health environments. Rory Kelleher of NVIDIA described the shift toward “Physical AI,” where agents increasingly contend with real-world constraints and lab automation rather than just text and images.

From isolated tools to orchestrating discovery

If smarter science has a practical shape, it is orchestration.

Brown illustrated this through a connected discovery sequence: moving from a 3D shape search at trillion-scale to ADMET profiling, docking and retrosynthetic planning. At SapioCon, this was mapped to an ecosystem of specialist capabilities, including OpenEye, Simulations Plus, CCDC GOLD and Elsevier Reaxys. Each step was initiated from the notebook, with outputs returned into the same validated record. The goal is not to pretend one platform can do everything. It is to keep best-fit methods usable in context so scientists spend more time on judgment and experimental design, less on stitching outputs together across disconnected tools.

Thank you to our partners and customers

Speakers and delegates at SapioCon 2026 were clear: the question of whether AI belongs in the lab is long past. The question now is where it adds the most value inside real workflows, how it does so without breaking governance or traceability and when teams will see measurable returns that hold up under scrutiny.

Cramer tied that direction back to Sapio’s mission of reducing suffering by getting solutions to the market faster. We are no longer just dreaming of smarter science. We are building the foundation it depends on.

Thank you to the 146 attendees from 24 customer organizations who joined us at the Renaissance Boston Seaport District to help define the future of smarter science. We also thank our speakers and partners from NVIDIA, Waters, Novo Nordisk, the Pistoia Alliance, Wellcome Sanger, AWS and Navignostics. Additional thanks to our sponsors and technology partners, including Leap Consulting Group, Astrix, Zifo, Workflow Informatics, Kalleid, ZS, Excelra, Cognizant, CCDC, OpenEye, Simulations Plus and Elsevier.

Key points

  • Scientists are already bypassing governed systems: more than three-quarters of respondents in a Sapio Sciences survey admitted to using unvetted AI tools outside official workflows to keep work moving.
  • The core bottleneck in R&D is coordination, not ambition, with industry estimates suggesting data scientists spend up to 80% of their time collecting and wrangling data rather than generating insights.
  • Sapio’s ELaiN converted a 28-page digital droplet PCR SOP into a structured, multi-tab experiment template in approximately seven minutes, a process that traditionally takes hours of manual configuration.
  • At Navignostics, a spatial single-cell proteomics platform built on operational data produced treatment outcomes three times more effective than standard care across a 120-patient study.
  • The emerging consensus among R&D leaders is that the data layer is the AI strategy: AI cannot fix processes that haven’t been standardized, and governance infrastructure must precede intelligence at scale.