Guide · the CEO explains
What is Zapier and what does it do? Zapier's CEO explains
Zapier connects the apps a business uses so work moves between them on its own. CEO Wade Foster explains what it does now, and where AI needs a human.
The short answer
Zapier is software that connects the apps a business already uses, so work moves between them without someone copying it by hand. Its basic unit is the Zap: when a trigger happens in one app, Zapier runs one or more actions in others, and it now also builds AI agents and lets AI tools like ChatGPT, Claude and Cursor act in more than 9,000 apps. In Nick's interview, co-founder and CEO Wade Foster walked through how it got there: simple app-to-app integrations in 2011, full workflows in 2016, and now AI. He also said things a product page would not. Many AI agents, he said, are really workflows under the hood. "These AI systems are not known for their reliability," he said, and they burn a lot of tokens. His advice is to use AI to design an automation, then run it as plain code. He has his own agents draft emails rather than send them, and said he might let them send only once he found he was not editing them 99 times out of 100. He also described the kind of agent he called really dangerous: one with private data, exposure to outside content, and the ability to send things out. Making all three safe together is, as far as he knows, unsolved.
See what Zapier does →Zapier sponsored this episode.
Step by step
It started as a link between two apps
Foster said Zapier began in 2011 as a simple way to connect tools like Dropbox, Evernote, Gmail and Trello. It was a trigger and an action: something happens in one app, and Zapier does something in another. That was the whole product.
Then it became full workflows
Users wanted more than one action, so in 2016 Zapier added full multi-step workflows. Foster's example was a lead from a Facebook ad: look up more about the account, save it to the right list in Salesforce, and text the sales rep right away if it is a hot lead.
Now it connects AI to your own tools and data ▶6:40
Foster calls this the third era. A plain chatbot gives a smart but generic answer, he said. Give the same AI your strategy, your customer profile and your account data, and it starts to answer like it knows the business. "The experience goes from vanilla to holy cow."
Zapier rebuilt itself around AI first ▶13:40
When GPT-4 came out, Zapier called the first code red in its history. It paused the whole company for a week to run a hackathon, and daily AI use among staff went from under 10% to over 50% in that one week in 2023.
Agents you describe in plain words
Nick found the agent builder easy for people who do not code. Foster said that was the goal: describe the agent you want in natural language, and the system builds it. Under the hood, he said, many agents are just workflows, some fully fixed and some with a bit of AI in certain steps.
Build with AI, then run it as code
Nick noticed the AI sometimes told him to skip the agent and use a plain Zap. Foster said that is on purpose. Talking with AI is the best way to design an automation, but once it works you want it to run the same way every time. The common mistake, he said, is letting AI make decisions at every step while it runs.
Plug Zapier into other AI tools ▶0:20
Zapier MCP and the Zapier SDK let tools like Claude Code and Cursor use Zapier's app connections. Foster runs a Saturday agent this way. It reads his week of emails, Slack threads, Cursor sessions and Granola meeting notes, then suggests half a dozen to a dozen automations. He usually has about half of them built.
Keep a human in the loop ▶44:02
Foster said the danger comes when one agent has private data, takes in content from outsiders, and can communicate outside the company. His rule is to give an agent only two of the three, and even then to keep a person checking its work at first.
“On the running side, run it with code.”
Wade Foster, in the episode
Key moments
- ▶0:20Foster's Saturday agent that studies how he worked all week
- ▶6:40Why AI with your business data feels different from plain ChatGPT
- ▶9:15Nick's AI advisory board spots what he missed about his own show
- ▶13:40Zapier's first code red, and a week-long company hackathon
- ▶22:55Why one person can now do every job in the chain
- ▶44:02The lethal trifecta: three things that make an agent risky
Daily AI use at Zapier jumped from under 10% to over 50% in one week. After GPT-4 launched in 2023, Zapier paused the company for a week and had everyone build with AI, not only engineers. Foster said that week moved daily use from under 10% of employees to north of 50%, at a time when most people were not using AI much yet.
Many AI agents are workflows underneath. Foster said a lot of what people call agents are step-by-step workflows: get a lead, enrich it, save it to Salesforce. Some are entirely fixed and some use a bit of AI for certain steps. That is why he thinks the workflow tools Zapier spent years building carried over so well into AI.
Use AI to build, and code to run. Foster said much of the value of AI is in designing the automation. Once it works, you want it to run exactly the same way every time. Nick noticed that Zapier's own AI sometimes told him to skip the agent and use a plain Zap, and Foster said that is deliberate.
His bar for unreviewed work is 99 out of 100. Foster's agents draft his emails but do not send them. He said he might let them send if he found he made no edits 99 times out of 100, because living with one mistake in 100 is worth the review time saved. For comparison, Zapier's agents guide suggests an agent over a Zap when 80% accuracy is genuinely sufficient, and a Zap when exactness matters.
An hour or two of follow-ups now takes under 10 minutes. Foster's days are full of meetings, which leave a long list of intro emails and status updates. An agent now reads his Granola transcripts and has the drafts waiting at the end of the day. He checks them and presses send.
Two of the three, never all three. Foster's rule for agent safety is to never combine private data, exposure to untrusted outside content, and the ability to communicate externally in one agent. He said whether all three can be made safe together is, as far as he knows, still unsolved.
How did Zapier go from app connector to AI platform?
Foster described Zapier in three eras. It began in 2011 as a simple trigger and action tool between apps like Dropbox, Gmail, Trello and Mailchimp. In 2016 it added full workflows, which let one event set off a chain of steps across many apps.
The third era is AI, and Foster argued the first two set it up. Integrations matter because AI is far more useful when it can see your real tools and data. Workflows matter because much of what an agent does is a series of steps. Today, he said, people opening Zapier mostly see Zaps, then Agents, then Zapier MCP and the Zapier SDK for connecting other AI tools.
The turn came with GPT-4. Foster said the short gap since 3.5, the better answers and the lower cost together convinced him it would change the whole software industry. Zapier rethought its roadmap and its operations around it.
What is the difference between a Zap, an agent and Zapier MCP?
A Zap is a fixed workflow: a trigger starts it and one or more actions follow, the same way every time. An agent works from plain-language instructions and can use live data and app actions, running on demand, on a schedule, or in the background, according to Zapier's guides. Zapier MCP lets an outside AI tool such as ChatGPT, Claude or Cursor reach the same app connections, with Zapier holding the logins.
Foster's practical advice ties them together. Talk to an AI to work out what you want. Once the steps are settled, run them as a plain workflow so they behave the same way every time. Some workflows, he said, keep a little AI for certain tasks inside them.
Where does Foster say AI should not run on its own?
Nick asked where humans need to stay in the loop. Foster pointed to the lethal trifecta: an agent with access to private data, exposure to content from outsiders, and the ability to communicate externally. He gave the example of an airline chatbot that, as he recalled it, was talked into a large refund. His rule is to give an agent only two of the three.
Even with two of the three, he recommends a human checking the work at first. His own agents write drafts, not sent emails. Over time he expects checks to look like factory quality control, where people sample some of the finished work and use statistics to decide the rest meets the bar.
He also sees AI changing jobs more than ending them. Many engineers no longer write code, he said, yet demand for engineers is growing, because the job was always solving problems, not typing code.
Method and sources: This page is written from the episode transcript and from Zapier's own pages listed in the sources.
Sources and references
Questions people ask
What can Zapier's AI features do?
On Zapier's own pages, the AI side includes Agents that work across your apps in the background, Chatbots that answer questions, Copilot, an assistant that helps you build workflows in plain language, and Zapier MCP, which lets AI tools like ChatGPT, Claude and Cursor take action in more than 9,000 apps. In the episode, Foster described what users notice: AI that can read a business's email, calendar, Slack and notes, and connect things across them that no single person had seen. He uses it himself to draft his end-of-day emails and Slack updates.
How do Zapier AI agents work?
You describe what you want in plain language and Zapier builds the agent. Zapier's guide says agents carry out multi-step work across your apps using instructions, live data sources, and app triggers and actions, running on demand, on a schedule, or when you are not around. Foster's example was leads coming in from Facebook: enrich each one, then route it to the right place. He added that many agents are really workflows under the hood, and that Zapier runs the fixed parts the same way every time because that is more reliable and costs less.
Is Zapier worth it?
This page cannot settle that for you, and Zapier sponsored the episode. Here is what came up. Foster said his end-of-day follow-ups used to take an hour, maybe two, and now take under 10 minutes because an agent drafts them. Nick said his own company became much more efficient once it started automating repetitive tasks with Zapier. Foster's suggestion is to look at the tasks you repeat every day, week and month and automate as much of them as you can. If you rarely move information between apps, there is less for it to do.
What is the Zapier MCP server?
Zapier MCP lets AI tools such as Claude, ChatGPT and Cursor take real action in your apps, with Zapier handling the logins and the permissions you set. You describe the goal, and the AI searches Zapier's apps and actions for the right tools. Zapier says it is included in every plan, and each tool call uses two tasks from your existing quota. Foster described it as a way to bring Zapier's integrations into tools like Claude Code and Cursor in a governed, safe way. He uses the related Zapier SDK with Cursor for his own weekly agent.
What is a Zap?
A Zap is an automated workflow. Zapier sums it up as a simple command: when this happens, do that. Every Zap has a trigger, the event that starts it, and one or more actions, the things it does next. Foster said Zaps are still most of what people see when they open Zapier today.
Who is Wade Foster?
Wade Foster is a co-founder and the CEO of Zapier. He grew up in Missouri, finished school with an engineering degree during the financial crisis, and took a marketing job at a small software startup in Columbia, Missouri. Seeing customers in Australia find that product, sign up and pay without ever talking to anyone is what made him want to start his own software company.
Do you need to know how to code to use Zapier?
No. Zapier's getting-started guide says no coding is required. In the episode, Nick said the agent builder suits people who are not engineers, and that tools like Claude Code can be more power than most people need. Foster said the aim is that you describe what you want in natural language and the system builds it.
Should you use an AI agent or a regular Zap?
Foster's answer is to use AI to design it, then run it as a plain workflow whenever the steps are the same every time. Letting AI make a decision at every step, he said, is where people go wrong. Zapier's own agents guide draws a similar line. It suggests a Zap when exactness matters, and an agent when 80% accuracy is genuinely sufficient.
What is the lethal trifecta for AI agents?
It is a framing Foster called the best he has heard. An agent becomes dangerous when it has three things at once: access to private data, exposure to content from outsiders, and the ability to communicate externally. His rule is to build agents with only two of the three. Whether all three can be combined safely is, as far as he knows, an unsolved problem.
What did Zapier do when GPT-4 came out?
It called a code red, its first ever. Foster said three things stood out: GPT-4 arrived only six months after 3.5, it was clearly smarter, and it cost less. Zapier changed its roadmap, bought AI tools for staff, and paused the whole company for a week to run a hackathon for everyone, not only engineers. Daily AI use went from under 10% of employees to over 50% that week.
How does Wade Foster use Zapier himself?
He connects Cursor to his work tools through the Zapier SDK. Every Saturday an agent reviews his week and suggests new automations. At the end of each day another agent reads his Granola meeting notes and drafts the follow-up emails and Slack updates he owes. He reviews the drafts and sends them himself. Work that took an hour or two now takes under 10 minutes.
Zapier sponsored this episode of the show. Nick chose the questions, and Wade's answers are his own.
Full episode: AI Knows Your Business Better Than You Do · All guides · Sponsor the show