Clay's New Workflows Feature Lets Its AI Agent Claygent Make Branching Routing Decisions on Live GTM Data

Clay shipped Workflows in open beta on August 11, an orchestration layer that runs automated GTM plays directly on top of a team’s live data instead of exporting records to a separate tool. The headline change is who — or what — makes the routing decisions: branching logic can be set with simple field rules, written as Python expressions, or handed to Claygent, Clay’s AI research agent, to judge case by case. A record entering a workflow can trigger off audience membership changes, a schedule, a webhook, or a cron job, then get routed through that branching logic before results write back into Audiences, Clay’s underlying data layer.

The build mechanics matter for how fast teams can actually stand this up: Workflows can be assembled through natural-language prompts via Clay’s Sculptor assistant, through CLI or API for coding agents, or on a visual canvas. Code nodes let users drop in custom Python for transforms and scoring. Clay also built in record-level observability — showing the exact path and decision a given record took — plus per-step cost transparency on credits and actions consumed, with no caps on rows or steps. It ships at no added cost to Enterprise, Growth, and Launch subscribers through the rest of 2026.

Clay’s own growth team already showed what this kind of AI-judged routing looks like running in production. In a companion post on Clay’s growth playbook, the company described “account agents” continuously monitoring roughly 100 tier-one target accounts for buying signals, stitching together Snowflake, Gong, and Salesforce data through Clay itself to drive expansion and churn-prevention work — the same branch-on-a-signal pattern Workflows now formalizes into a general-purpose product.

For outbound and RevOps teams, the shift is from writing static if-this-then-that automation to letting an agent make the routing call and showing its work at the record level. That auditability — knowing exactly why Claygent sent one account down one path and another down a different one — is what turns “the AI decided” from a liability into something a RevOps lead can actually defend to a VP asking why a deal got skipped.