Workflows First for Vertical AI
Levi Lian, CEO of Raycaster, on why vertical AI is workflows first & chat last
This is really a statement about where the product moat lives. Standard MCP endpoints from Veeva or IQVIA would make integration cheaper and cleaner, but they would not solve the hard part of life sciences work, which is mapping each company’s document types, approval rules, templates, and evidence chains into an AI system that can actually draft, review, and route work inside regulated workflows.
-
Veeva and IQVIA already act as the compliant systems of record. They store trial, quality, and regulatory documents, but the missing work is still the in between layer, what is incomplete, what conflicts, what should be drafted next, and who needs to review it. MCP helps connect to those systems, but it does not create that judgment layer.
-
Raycaster is built as a system of intelligence and action on top of those records. In practice that means reading specs, methods, batch records, and Module 3 sections, finding page level gaps, proposing edits, and sending work to the right owner. The integration pipe matters, but the value is in the workflow logic and domain ontology riding through that pipe.
-
This is why the company uses a forward deployed model early. Engineers map a biotech customer’s actual handoffs across Veeva, IQVIA, SharePoint, and LIMS, then codify review roles, acceptance criteria, and edge cases into reusable modules. Cleaner APIs reduce services effort, but they also let the product team spend more time productizing these modules faster.
As more core life sciences platforms expose richer APIs and eventually MCP style interfaces, the integration layer will commoditize. The winners will be the companies that turn that easier connectivity into trusted, audit ready workflow products, with domain specific checks, reusable process maps, and accumulated evaluation data that improves each document cycle.