Getting started
The plumbing tax
Replacing a five-minute Zapier scenario with your own backend costs a server, a deployment, auth, and a webhook endpoint. You pay it once, and then a new workflow is close to one prompt.
The plumbing tax is everything you have to build before the first line of business logic exists. It is the reason engineers reached for Zapier for years, even when they could have written the script in twenty minutes.
Take a simple automation. A new Stripe payment should land in an Airtable base for reporting. In Zapier that is five minutes. Pick the two integrations, click through OAuth, add three nodes, done. The first time you build the same thing yourself, it is not five minutes.
The cost of the first one
Before that Stripe payment reaches Airtable, you need:
- A server to run on
- A backend framework for the endpoints, such as FastAPI
- A deployment, and a strategy for deploying again next week
- CI/CD, so deploying is not a manual job on someone's laptop
- Authentication for the caller, through API keys or OAuth
- Secret handling, so keys never sit in the repository
- The webhook endpoint itself, with validation
- A way to test and debug the whole path locally
None of that is the automation. All of it has to exist first.
Why it is affordable now
AI coding agents made this work cheap to build. Plumbing is exactly the kind of well-understood, heavily documented code an agent writes well, so automating things that were never worth automating now pays off. It also stopped being something only enterprises with a platform team can run.
You pay it once
Once the foundation exists, about 99% of the plumbing is done forever. The server, the deploy, the auth, the queue, the event store, the retry logic, all of it is already there. Adding a new workflow on top is close to one prompt.
The first workflow
Every workflow after
The mistake to avoid
If you never centralize, you pay the plumbing tax again for every automation. Different server, different backend, different language, different database. That is how an automation setup turns into a mess nobody can debug.
The same problem shows up inside the low-code tools, just later. Every scenario is cheap on its own, so you keep adding them. A year in you have sixty, and when one breaks or an API version is retired, finding the broken one is hard.
Set the standard once and you have an automation backend. Then you can go back and build the workflows that were not worth it before, because everything is already connected.
Automation types
Every job in an automation platform either reacts to an event or runs at a scheduled interval. Webhooks handle the first, Celery Beat handles the second, and both end up in the same event ledger.
AI transformation
Bottom up means giving individual people AI harnesses and skills. Top down means one company backend that runs work in the background. Both matter, and only one of them is getting attention.