Guide · made with Zapier
Is Zapier worth it? A cost test from Nick's own AI workflows
Zapier is worth it for steps with a right answer you can write as a rule. Pay for a model only where judgment is needed. Here is how to tell.
The short answer
Is Zapier worth it? For the steps in a workflow that have a right answer you can write down as a rule, yes, and in Nick's example that was most of them. The model is worth paying for only on the steps that need judgment. Nick makes this case in a video sponsored by Zapier. His point is that much of an AI bill goes to work that does not need AI at all, with a model predicting its way through if-then statements and charging for it. That pattern is what people call token maxing. The fix is a test, not a product. Break one workflow into its real steps and ask of each one: is there a right answer here? A trigger, a text match, a database read, and formatting all pass that test. They can run as plain automation for zero tokens, and Nick says that kind of step is more reliable than a model would be. Summarizing, drafting, and calls the rules cannot cover are where a model earns its cost. In his competitor-monitoring example, one of six steps needed the model. Zapier modeled 24 common agent workflows this way and found 71% average savings. Zapier notes that figure comes from modeled workflows, not live customer ones, so treat it as a direction, not a promise.
Get the Zapier framework guide →Built with Zapier, the sponsor of this video.
Step by step
Start with what the AI spend is achieving ▶1:28
Nick's first move is to change the question. Do not ask how to spend less on AI. Ask what the AI spend is actually achieving. Once that is answered, he says, the cuts start to make themselves.
Write one workflow down, step by step ▶4:23
Pick a workflow you run now, or one you are about to build this week. List every step in order, from what starts it to where the result lands. Nick's own example had six steps once it was written out.
Ask of each step: is there a right answer? ▶3:25
If a step has a right answer you can write down as a rule, it is an if-then. That is Nick's first rule: default to deterministic. Those steps cost zero tokens in plain automation, and he says they will be more reliable than a model. This is the part of the bill where an automation tool like Zapier pays for itself.
Pay for the model only where judgment lives ▶3:38
Nick's second rule is to spend tokens only on judgment. He names summarizing, classifying things that are unclear, drafting, and deciding when the rules do not really apply. "That's where a model earns its cost." If a step is none of those, it does not need one.
Do not add a second agent to check the first ▶0:58
A common pattern is to hand an agent work that plain automation could do, then build a second agent to check its work. Nick points out that now you are paying twice for one outcome. If a step has a right answer, a rule can do it without a checker.
Keep the model swappable ▶3:51
Nick's third rule is to keep model options open and not hardwire one provider. Price and performance shift every couple of months. If changing models is a project instead of a dropdown, you lose an easy win.
Ask your AI model which steps could be deterministic ▶4:35
Nick ends with a check anyone can run. Paste the workflow into your AI model and ask which steps could be deterministic. His guess is that fewer steps need a model than you think.
“If the task has a right answer that you can write down as a rule, it's an if-then.”
Nick Standlea, in the video
Key moments
- ▶0:00Most of the bill is work that does not need AI
- ▶0:15Sponsor note: Zapier
- ▶0:46What token maxing means
- ▶0:58Paying twice for one outcome
- ▶1:11Why models miss on rule-based tasks
- ▶1:28The better question about AI spend
- ▶2:54Six steps, one needed the model
- ▶2:59Zapier's 24 workflows and 71% average savings
- ▶3:22The three rules for value maxing
- ▶4:23Before you build an agent this week
Worth it is a question about steps, not tools. Nick never reviews Zapier as a product. He sorts work into two kinds: steps with a right answer, and steps that need judgment. Automation is worth it for the first kind. A model is worth it for the second.
Most steps have a right answer. In Nick's competitor-monitoring example, the trigger, the competitor match, the CRM read, the cleanup, and the post to Slack all had right answers. Only the step that reads the news and says what it means needed a model. That is one of six.
Zapier's modeled average was 71%. Zapier modeled 24 common agent workflows by routing the steps with a right answer away from the model. The average savings was 71%, and per-task savings ranged from 54% to 90%. Zapier says the study is modeled, not drawn from live customer workflows.
Rules are also more reliable. Cost is half the case. Nick notes that a model predicts a likely answer rather than running fixed logic, so it will sometimes miss on a rule-based task. A rule does the same thing every time, for zero tokens.
Checking an agent with another agent doubles the bill. Teams that hand an agent rule-based work often add a second agent to check it. Nick's point is that this pays twice for one outcome, when a rule would not need checking at all.
How do you decide if Zapier is worth it for your workflow?
Write the workflow out in plain steps. Next to each one, write either rule or judgment. A step is a rule if you could write down the right answer ahead of time: when this happens, do that. A step is judgment if it means summarizing, drafting, sorting something unclear, or deciding when the rules do not fit.
Count the rules. Each one is a step you are paying a model to guess at today if an agent runs the whole workflow. Those are the steps where an automation tool earns its place, because they run for zero tokens. The judgment steps are where you keep the model, and Nick is happy to pay for those.
If your list comes back mostly judgment, the savings will be smaller, and Nick says as much: results depend on how complex your workflows are and which models you use. If it comes back mostly rules, as his example did, the case for moving those steps to plain automation is strong.
Method and sources: Claims about the video come from its YouTube transcript, with timestamps taken from transcript start times; the 24-workflow study, the 54% to 90% range, and its modeled-only caveat come from Zapier's valuemaxxing page, and the AI feature list comes from Zapier's agents guide, all checked when this page was drafted.
Sources and references
- Zapier: Valuemaxxing (modeled study of 24 AI agent workflows)
- Zapier Agents
- Zapier blog: Zapier Agents guide
Related guides
Questions people ask
Is Zapier worth it?
The answer from Nick's video depends on the step, not the tool. For steps with a right answer you can write down as a rule, like a trigger, a text match, a database read, or formatting, plain automation does the job for zero tokens and does it more reliably than a model. In his example that was five of six steps. The model is worth paying for only on the step that needs judgment. The video is sponsored by Zapier, so weigh that, and run the step test on your own workflow before deciding.
What can Zapier's AI features do?
Nick's video does not demo them; it is about deciding where AI belongs. Zapier's own agents guide says its agents are built from plain-language instructions, can be started by events in other apps, can use live data sources like Google Drive or Notion, can search the web, can run as steps inside a Zap, and can call other agents. There is also an activity log that flags runs needing your input. Nick's rule still applies: use the AI parts for judgment and let ordinary automation handle the steps with a right answer.
Which AI tools save the most time?
Nick did not rank tools, and this page will not invent a ranking. His video is about cost and reliability. His claim is that the savings come from taking model calls out of steps that never needed judgment, whichever tools you use. Zapier modeled 24 common agent workflows that way and found 71% average savings, a modeled cost figure, not a time study.
What are the best AI automation tools in 2026?
Nick tested one approach in a sponsored video, not a market. He walked one workflow and pointed to Zapier's modeled study; he did not compare products. What he offers instead is a test you can apply to any tool. Can it run the steps with a right answer as plain automation, for zero tokens? Does it let you spend model calls only on judgment? And can you switch models from a dropdown rather than rebuilding? A tool that does those three things fits his rules.
Is the 71% savings a real-world result?
It is a modeled result. Zapier's page says it modeled 24 AI agent workflows in an internal study to sort which steps needed AI and which did not, and that it is not based on live customer workflows. Per-task savings in the study ranged from 54% to 90%. Nick adds that your mileage will depend on how complex your workflows are and which models you use, but the direction holds.
Why not just let an AI agent do every step?
Nick's video gives a cost reason and a reliability reason. Cost: every step an agent does is a model call, and much of that is work that needs no AI. Reliability: a model predicts a likely answer instead of running fixed logic, so on a rule-based task it will sometimes miss, and it misses at a premium price.
What is token maxing?
It is the pattern of throwing the model at everything and assuming the value shows up later. Nick's point is that the value does not always show up. The opposite approach, value maxing, spends tokens only where judgment is needed.
Is this video sponsored?
Yes. Nick says the video is sponsored by Zapier and that it is a topic he would have made anyway, because he keeps seeing people and businesses run into the same problem.
Zapier sponsored this video. Nick chose what to build and what to say.