Workflow automation
Workflow automation is a fixed sequence of steps a person designs in advance, run by software instead of by hand, that can't deviate from the branches it was built with, as distinct from a tool that does one thing on request or an agent that decides its own next step.
A tool does one thing when asked: a lookup, a single API call, nothing decided. Workflow automation is one level up. It chains multiple tools in a sequence a person wrote ahead of time, and it runs that sequence without a person clicking through each step by hand. An enrichment waterfall is the clearest example: try one data source, and if it comes back empty, try a second, then a third, logging what came back at each stop. Every branch of that sequence was decided in advance. Nothing in it improvises.
Why the line between workflow and agent matters
Workflow automation and an AI agent get used as if they're the same thing at different levels of sophistication. They're not. The difference isn't how many steps run or how advanced the model behind them is. It's who decided the branch. A workflow follows branches a person specified in advance, however many steps it has. An agent decides what to do next based on what it just observed, without a person approving that specific branch first, the way a system watching inbound replies might decide on its own whether to follow up, skip a lead, or escalate to a rep.
Most tools marketed as AI agents are workflows with a language model writing the copy inside one of the steps. That's not a failure, workflow automation is often exactly the right tool for the job, but it's a different claim than an agent, and the difference matters the day something breaks and a team has to trace which step made the call.
In practice
Most GTM engineering work that gets called automation is workflow automation specifically: monitoring signals, running an enrichment waterfall, routing a scored lead into a sequence. It's not glamorous, and it doesn't need to decide anything a person hasn't already specified. It just needs to run the specified sequence without someone doing it by hand every time.
What people get wrong
People assume adding more steps or a fancier model makes a workflow 'more agentic.' It doesn't. A multi-step process a person fully specified in advance is still workflow automation, however many steps it has. The real test is whether any single branch in it gets decided live, by the system, rather than by whoever built the sequence.
