Waters Opens Its LC-MS Software to AI Agents and Joins Anthropic’s Model Hardware Standard 🔗

On 13 October, Waters launched AI Link, which brings the Model Context Protocol to waters_connect, its LC-MS informatics platform. Agents working with Claude, including Claude Science, or other large language models can now create methods, run acquisitions and process results without custom integration, and Waters joined Anthropic’s Model Hardware Standard research programme. The regulated angle matters most: work started in waters_connect can carry its audit trail into Empower, Waters’ compliant system of record for QC. General availability is planned for November. Waters says AI-directed lab work is already a fast-growing source of demand for its instruments and consumables, because every hypothesis a model generates becomes a sample that has to be measured.

Benchling Launches Science Foundries: Teams of Agents Running Whole R&D Workflows 🔗

On 15 October, Benchling launched Science Foundries, where R&D teams pick a goal, such as producing proteins or optimising a lead, and decide which steps agents run and where scientists approve. In one antibody example, separate agents design sequences, place cloning orders with a CRO, dispatch work to lab automation and update training datasets with the results. Benchling says agents are the fastest-growing capability it has ever released. With more than 1,300 customers, it is positioning its R&D data platform as the coordination layer above instruments, CROs and generative models.

Biohub, DOE, NIH and Big Tech Commit $1.8 Billion to AI-Ready Biological Data 🔗

On 7 October, Biohub expanded its Virtual Biology Initiative into what partners call the largest coordinated commitment to AI-ready biological data to date. The DOE will invest more than $500 million over five years, the NIH contributes datasets from more than $500 million of prior federal investment, and Google DeepMind, Isomorphic Labs and Meta are investing $300 million, on top of Biohub’s founding $500 million. The Broad Institute, the Human Cell Atlas and the Wellcome Sanger Institute are among the scientific partners. The result will be an open resource on how cells respond to interventions. Next to last issue’s federated co-folding results, the market is now funding two data strategies in parallel: proprietary data pooled without being moved, and public data generated at a scale no single company could afford.

Agilent Buys Proscia for $240 Million 🔗
Agilent is acquiring digital pathology software company Proscia in an all-cash deal worth $240 million, with up to $30 million more in performance milestones. It gets Concentriq, a platform used by more than 4,000 laboratories, hospitals and pharma companies to manage pathology images, data and workflows, with support for third-party AI applications. The two companies have had a distribution agreement since 2022.

The signal: In May, Roche bought PathAI. Now Agilent, which sells the stains, reagents and instruments that produce tissue data, is buying the vendor-neutral layer where that data is managed and analysed by AI. The companies that generate the physical sample want to own the software on top of it, and for independent pathology AI developers, distribution increasingly runs through platforms owned by large diagnostics suppliers.

AI Discovery Platforms Sign Pharma Deals While Iambic Goes Public 🔗

On 15 October, insitro signed AstraZeneca for MASH and cardiomyopathies: $40 million upfront and up to roughly $1.46 billion in milestones, with insitro’s Virtual Human platform taking targets to development candidate and AstraZeneca running the clinic. Two days earlier, Zealand Pharma signed Generate Biomedicines for AI-designed metabolic disease proteins, with $30 million upfront and nearly $1.2 billion in milestones. And Iambic raised $150 million in its IPO at $16 per share.

The signal: AI discovery deals are settling into a recognisable shape: a $30 to 40 million upfront, a billion-dollar milestone tail, the platform leading discovery up to a candidate and pharma taking the clinic. Pharma is paying for platforms with experimental results behind them. Iambic adds a public market reference point, though BioPharma Dive notes that several AI drug companies have struggled after listing, so clinical readouts will matter more than the platform story.

IQVIA Makes Veeva Its Clinical Operations Standard and Launches Predictive Models 🔗

On 13 October, IQVIA selected Veeva CTMS, eTMF, Study Startup and Site Connect as enterprise standard applications across its clinical development studies, extending the partnership the former rivals started in August 2025. In the same fortnight, IQVIA launched IQVIA Life Science Models, built using NVIDIA Nemotron on AWS, for study simulation, protocol optimisation and synthetic controls. In an initial evaluation across 50 trials and 270 endpoints, IQVIA reports that 85% of predictions were on target or clinically close.

The signal: The world’s largest CRO is standardising on a competitor’s system of record and putting its own effort into data and prediction. Core trial operations software is turning into shared infrastructure, while differentiation moves to the layer that predicts how a trial will perform before it starts. Sponsors comparing CROs should ask how those prediction claims were validated.