n8n is what we use for AI workflows. One run counts as one execution however many steps it has, it's free to self-host, and Claude can write the workflow JSON for you. Zapier is the easiest to start and has the most integrations, but it charges per task (each step), which adds up on multi-step AI flows. Make's free tier gives 1,000 credits a month. Pipedream now targets developers embedding integrations in their own apps. OpenAI's Agent Builder shuts down on November 30, 2026.
Workflow tools meter in different units: n8n counts whole executions, Zapier counts tasks (each action step), Make counts credits and Pipedream counts compute time. The same ten-step AI flow can cost one execution on n8n and ten tasks on Zapier.
Ryz Labs runs an n8n workflow that researches companies, finds contacts, writes a custom email with OpenAI and saves a draft to Gmail. Sam was candid about getting started:
while it's theoretically low code or no code, to get started, there I think there is a learning curve.
— Sam Nadler, Episode 17, at 11:29
The shortcut is to have Claude write the workflow and import the JSON, but that only gets you most of the way:
the last 10% of connecting it all, which can be at times challenging
— Sam Nadler, Episode 17, at 15:34
Two practical limits showed up in Jordan's live run. He built in wait periods "so that we don't, like, spam any of our APIs" (at 13:59), and "it'll take a few minutes to run this entire task" (at 14:46). The upside: n8n shows "how many tokens, like, each job uses as it calls the different OpenAI APIs" (at 15:09).
I've checked out the chat GPT workflow tool, agent workflow tool… It's very, very rudimentary, I would say, now. It definitely doesn't have as many features that [n8n] does.
— Jordan Metzner, Episode 17, at 16:26
OpenAI is now shutting it down (see the table).
Jordan's always-on agent setup (OpenClaw on a cloud server) was "very much like cron job type or task based focus", (at 20:14), and in early 2026 he found "the value is still pretty low" (at 20:01). Desktop agents have the opposite problem:
as long as you sit next to them, they'll do it. But, like, if you don't sit next to them anymore, they will stop working.
— Jordan Metzner, Episode 57, at 08:52
His fix was an agent with its own server, which "can do these kind of cron job flows and workflows… and stay maintained and logged in" (at 09:12). For simpler schedules, Sam ran a Supabase edge function on a GitHub cron job (at 06:59).
We have no build of our own on Zapier, Make or Pipedream. Jordan has only called n8n "probably not that much different from Zapier and Make" in concept (at 12:14). They appear in the price table only.
Breeze CEO Karim Malhas said early n8n "only had one project, and you have everything inside of it", which didn't work for the enterprise he was at (at 02:55).
n8n's Community Edition is free to self-host. On November 21, 2026, n8n Cloud Starter was 20 euros a month billed annually for 2,500 workflow executions, and Pro was 50 euros a month billed annually for 10,000 executions, with no overages (runs stop at the cap).
Usually, for multi-step AI workflows. n8n counts a whole workflow run as one execution with unlimited steps, while Zapier counts tasks (each action step). Zapier Free includes 100 tasks a month and Professional starts at $29.99 a month ($19.99 billed annually) for 750 tasks.
Make's free plan includes 1,000 credits a month. We couldn't verify paid-plan prices on November 21, 2026 because Make's pricing page blocked automated access, so check make.com/pricing directly.
OpenAI's documentation says Agent Builder is deprecated and scheduled to shut down on November 30, 2026. When we tried it in 2025, Jordan called it very rudimentary compared to n8n.
Yes. On the show we asked Claude to write an n8n workflow, downloaded the JSON and imported it into n8n in seconds. Sam's estimate is that Claude gets you about 90% of the way, and the last 10%, connecting accounts and APIs, can be challenging.