Workflows First Chat Last
Levi Lian, CEO of Raycaster, on why vertical AI is workflows first & chat last
This reveals that the winning vertical AI products are becoming software for specific jobs, not smarter chat boxes. In regulated work, users need the system to open the right documents, check them against templates and rules, propose edits with page linked support, and send them to the right reviewer. That is why Raycaster is building click to draft and click to review modules for tech transfer, QA, and Module 3 work, instead of leading with open ended chat.
-
Harvey is the clearest proof that chat alone does not hold up. Its positioning shifted away from a proprietary legal model and toward pre configured agent workflows, plus high touch implementation and ex lawyers in customer success, because adoption depends on changing how firms actually do work day to day.
-
Hebbia points to the same pattern from the finance side. The durable product is a control surface that lets an analyst run retrieval across many documents in a structured table or workbook, so the user is steering a repeatable process, not asking a general purpose assistant to think from scratch every time.
-
In life sciences, this matters even more because the core artifacts are regulated operating documents, not just research notes. Raycaster sits on top of systems like Veeva and IQVIA, wiring in templates, permissions, approval roles, and evaluator feedback so edits can be verified and routed instead of living as unverifiable chatbot output.
The next wave of vertical AI will look more like packaged workstations inside each regulated function. The companies that win will own the workflow map, the acceptance tests, and the human review loop. Chat will remain in the interface, but mainly as a convenience layer on top of systems built to produce auditable work products at scale.