I was talking with a founder recently who's dealing with something pretty intense. He's building a tool that handles finance and billing in a highly regulated industry. We're talking healthcare insurance billing rules, coordinating different physician payment structures, and industries with historically thin margins. The kind of environment where an error in billing or even the timing of when billing happens can create real consequences.

He asked me something that really stuck with me: "I know I need AI because there's too much nuance between customers and too many variables in billing rules for a simple rule based engine. But do I actually need to build an agent for this, or will a workflow solution work?"

This is the question I'm hearing often. And it's a good one.

Here's what I told him.

Start with the workflow

He'd already identified something critical. Most of his product's workflow is deterministic. Some back office function kicks things off, then there's a specific point where AI handles the nuance, and after that it's back to a predictable sequence of steps to get to the end result.

When that's your situation, a workflow solution is almost always better. Here's why.

Testing becomes straightforward. Your eval system stays manageable. You can actually understand what's happening when something goes wrong.

Compare that to starting with agents. When you're debugging and you've got complicated multi-turn actions with several tools, your logs and traces get messy fast. And it only gets worse the longer your system runs. The more your product grows, the more complicated everything becomes.

You can always layer on agentic functionality later. When specific use cases call for it. But until then, the workflow will dramatically simplify your life.

The right tool for the job

Agents are powerful. The excitement around them is real and justified. But like any tool, they shine in specific situations. Choosing the right tool for the job is one of the harder problems builders are facing right now.