Pick by who will run it and how often it will run. Zapier is easiest for a non-technical team. n8n gives you more control and more volume for the money. Custom agents are for hard, high-volume work the platforms cannot do.
How do they compare?
- Zapier. Hosted for you, the biggest library of app connections, the easiest to learn. Billed by task volume, so it gets expensive as volume grows. Best for a non-technical team with a few simple automations.
- Make. Hosted and visual, with more control over branching than Zapier. Billed by operations. A middle step between Zapier and n8n.
- n8n. Hosted for you or run on a server you rent. Lets you drop into code when you need to, and handles AI steps well. Takes more skill to set up. Best when volume is high or you want the workflows on your own server.
- Custom agents. Code written for your process, calling AI models directly. The most control and the most responsibility. Best when the work needs judgment on documents or images at high volume.
Platform pricing changes often. Check the current tiers on their price pages before you decide.
What does a platform look like when it is working?
At Quartz Importers MN, where I hold the fractional CTO seat, a small sales team runs on 49 automated workflows every day. Those workflows have logged 39,645 production executions at a 0.2 percent failure rate. The automations do the routine work, and a person gets an alert when something fails.
When do I need custom agents instead?
When the work looks more like reading than moving. Platforms are good at "when this happens, do that." They get awkward when every item needs judgment: reading a scanned document, sorting it, pulling the right numbers out, deciding where it goes.
Before my current fractional seats, I built a system that sorts 100,000 documents a day at better than 95 percent accuracy. Every document needed a judgment call, so it had to be custom.
What should I not do?
- Do not pick the tool first. Map the process, then pick. Pick the tool first and the business ends up shaped around it.
- Do not let one person build fifty automations with no notes. When that person leaves, nobody knows what runs.
- Do not build a custom agent for something Zapier does in an afternoon. You maintain custom code forever.
How do I decide for my business?
Count the runs per month and name who will fix it when it breaks. Those two answers pick the tool most of the time. What an AI automation project costs covers the build and the monthly bill, and custom AI versus off-the-shelf tools covers the bigger version of the same choice. The AI build page says what I build, and you can describe your process on the start form.
Answered by Jared Lukes, Minneapolis. Twenty-five years in brand, UX, and technology, currently holding fractional AI and technology seats at three companies. About Jared.