Benchling puts agent teams inside R&D workflows: in a recombinant protein foundry, agents draft the design, place the order, run instruments and write the report
Synopsis
At Benchtalk 2026, Benchling launched Science Foundries, where teams of AI agents work alongside scientists across a full R&D workflow rather than speeding up a single step. In a recombinant protein foundry, one agent drafts the cloning design, another orders from a CRO and calls Benchling Automation to run instruments, and an Analysis Agent writes the report, with a scientist signing off at every critical decision. The product is in preview for customers, and the company is building foundries with customers across target validation, screening, method development and biologics production.
Interpretation
Benchling introduced Science Foundries, shifting R&D workflows from agents automating a specific task to teams of agents working alongside scientists through a whole process, so scientists, agents, models and data act as one system. Most AI so far has delivered incremental gains, with agents automating a specific task to make each step faster; the company argues the larger opportunity lies in redesigning workflows entirely. The recombinant protein foundry is the concrete example: a Benchling Agent drafts the cloning design, another orders from a CRO and calls on Benchling Automation to run the instruments, and an Analysis Agent writes the report, with a scientist signing off at every critical decision.
The agents draw on Model Hub, which makes structure prediction, property prediction and company models callable in the same system as the data. Models available in Model Hub include structure prediction models AlphaFold 2, Chai-1, Boltz-2, OpenFold 2, OpenFold 3 and Protenix, ADMET models from Inductive Bio, Lilly TuneLab proprietary developability models trained on decades of Eli Lilly R&D, the open source DeepSP model, and Boltz through a secure API. Every model in Model Hub can be called by a Science Foundry as one step in a larger workflow; support for running your own models is coming soon.
Perspective
The demonstrated scope is the product preview Benchling announced at Benchtalk 2026: Science Foundries are in preview for customers, the recombinant protein foundry is the example flow given, and the company is building foundries with customers across target validation, screening, method development and biologics production.
The company states that most prior AI delivered only incremental gains and places the larger opportunity in redesigning workflows, which is its judgment of the opportunity rather than a controlled measurement; support for running your own models is marked coming soon, and antibody-drug conjugate and peptide support is only in preview.