To Code or Not to Code

My LinkedIn feed used to be full of screenshots showing complex workflows in Make, n8n, Pipedream—all those "no code" builders where you click and drag automations like "When an email arrives with this label, use GPT-5 to summarize it and send it to this WhatsApp group."

I'm torn. For straightforward workflows, they're a beautiful entry into process automation, especially for people without programming skills. But when I see a screenshot with several dozen nodes and branches? Maintenance nightmare. Good luck changing the business logic. Good luck testing edge cases. Good luck debugging when something breaks. What starts simple becomes a ball of chaos.

They work well for simple integrations that let your software tools talk to each other in novel ways. Infused with AI to extract or reformat information, they can be useful. But as soon as you're handling complex business logic, write code.

There's another problem: because these builders are graphical, not text-based, you can't use coding agents to help you. Yes, they all come with their own "Agentic Workflow Builder AI" where you describe what you want and it builds it for you. But you don't get Claude Code or Codex. And whatever you build is locked to that platform. Want to switch from n8n to Pipedream? Good luck. Your Python code runs just as happily on AWS as on Azure or Google Cloud.

If these tools offer a quick path to an initial product for a client whose business case doesn't warrant custom development, I'll use them. Otherwise, plain code wins.