Internal Automation (Heaven) in 2026
I was about to build my own automation platform. Let that sink in for a moment. A software engineer, mass-procrastinating on his actual automations by planning to build the tool to run them. Even in times of "the agent will do the code", this is a bit too far.
Let me tell you how I got there - and why I didn't have to.
The migration trail ¶
It started with IFTTT. I had a few personal automations running there, the most important one involving Spotify. Then IFTTT broke the Spotify integration one too many times. I moved to make.com. That worked fine for a while, but is way too expensive for the purpose.
So when I finally wanted to add more automations - including some with agentic flows - make.com wasn't cutting it anymore. The natural next step seemed to be N8N, self-hosted on a VPS. Open source, pretty UI, big community.
Here's where things went wrong. N8N's whole identity is "no-code." And for people who don't code, that's probably great. But for a developer, no-code doesn't simplify things - it gets in the way. Any slightly more sophisticated logic led me to copying a JavaScript snippet into a text input field.
It made my toenails curl up - call it craftsmanship reflex. Pasting code into a text area when you have a Git repo feels fundamentally wrong.
What I actually wanted ¶
Once I stepped back, the requirements were pretty clear:
- Code lives in a Git repo, written in a proper editor (or by a coding agent)
- A platform executes that code as workflow nodes
- Nice UI for orchestration, scheduling, and observability
- Pre-built OAuth connectors for services like Spotify, Google, Slack - because that's the tedious part nobody wants to build from scratch
- Everything else, the actual logic, is just code - trivial to write (especially with a coding agent)
That's it. Not a no-code platform. A code-first platform with infrastructure around it.
I was genuinely about to start building this. I had the architecture sketched out in my head. Luckily, I opted for a quick round of research first instead: does this already exist?
Enter windmill.dev ¶
It does. It's called Windmill, and it does exactly what I had in mind. To be exact: it does more, better!
Windmill is a code-first automation platform. You write scripts in TypeScript, Python, Go, or Bash. Those scripts become the nodes in your workflows. Everything lives in a Git repo. The platform handles execution, scheduling, retries, error handling, and gives you a proper observability UI on top.
But here's what pushed it from "good" to "I'm genuinely excited about this":
- CLI and agent integration. Windmill ships a CLI that lets you create, test, and push workflows from the terminal. It even has dedicated skills for coding agents.
- OAuth connectors. Dozens of pre-built integrations. Spotify, Google, Slack, GitHub - the boring OAuth dance is handled for you. That alone saved me hours.
- Migration was trivial. I took my N8N Spotify workflow, exported it as JSON, dropped it into Claude Code, and it recreated it in Windmill within minutes.
My new workflows were equally quick to build. Describe the automation, let the agent write the scripts, push, schedule, done. Claude Code can even check observability stats and debug the scripts' live executions directly through the CLI.
The actual insight ¶
Here's what I think many teams are missing right now: the combination of a code-first automation platform like Windmill and a coding agent is absurdly powerful. Not just for personal Spotify playlists - for company-internal automation, internal tools, scheduled jobs, data pipelines, anything that currently lives in a Notion doc titled "we should automate this someday."
No-code tools solve the wrong problem for developers. We don't need drag-and-drop abstractions that hide the code. We need proper infrastructure that runs our code with observability, scheduling, and connectors - while staying out of the way.
Windmill + a coding agent is, in my opinion, the killer stack for internal automation in 2026. I can only recommend giving it a serious look.
