Zapier is a workflow automation platform that enables users to connect applications, automate business processes, and build AI-powered workflows without requiring coding expertise.
Zapier began by helping non-technical users connect apps without coding. How has the company’s vision evolved since its founding in 2011?
You can describe it in a handful of phases. Phase one was what you described: helping non-technical folks build integrations with the business applications they were using day in and day out, such as Salesforce, Dropbox, and Stripe.
The second phase was expanding into workflow automation. It became clear to us that, as valuable as integration was, workflow automation was much more valuable. Then we built a broader automation platform with tables and interfaces as a no-code application builder. Today, we are in the fourth phase, which is still about making automation work for everyone, but now with AI and agentic capabilities. That lets you build automations through natural language and bring inference into workflows, enabling use cases around summarization, decision-making, categorization, and more.
How are customers adopting agentic AI solutions in practice, and how comfortable are they with autonomous systems?
It depends on the use case. Lower-stakes use cases are where a lot of experimentation happens today, as people learn and get comfortable with how the technology works. As they become comfortable, they can figure out how to deploy it in higher-stakes, mission-critical workflows.
One thing we have found consistently is that building workflows with natural language is really powerful. Most people do not think like engineers or break systems down into step one, step two, step three. If you have a conversation with an agent, it can extract that logic from someone’s brain. But for mission-critical systems, there is still a lot of power in deterministic workflows because they provide reliability and cost advantages. The most savvy customers are using just enough agentic capability to run the workflow the way they want.
How does Zapier help customers move beyond experimentation with AI workflows?
Where Zapier really benefits customers is that our agentic builder can break down what you are trying to build, encode it into code, and run it deterministically, or as deterministically as possible.
As an end user, you do not want to think through whether something should be inference or determinism. That becomes complicated and jargony. You just want it to work. Our system can compile the workflow down to the ideal way to run it and then run it the same way every single time. That is where you get a lot of efficiency from running something on Zapier.
Can Zapier help close the AI adoption gap between large enterprises and smaller organizations?
There is a fair amount of learning curve right now in building these tools. I think of it like building a website in the 2000s, when you might hire an agency for a million dollars to build one. Today, that sounds absurd because you can use any number of site builders or even vibe code it. Building agentic systems is on a similar curve.
Our aspiration is to build tools that lower the barrier to entry. To me, the power and promise of AI is that it can onboard you to itself. If you can get people into that learning loop, it is incredibly powerful for self-paced learning because the technology can meet people where they are.
What is the biggest bottleneck preventing broader AI adoption today?
Hands down, the most common bottleneck is probably not what you expect. People simply do not know what to automate. They do not know where to start.
You see people online bragging about the agentic systems they have built, but if you ask what they actually built, the answer is often that they built their workbench. It is like a carpenter buying all the tools and setting up a beautiful shop, but not making a chair or a table. That is the biggest bottleneck right now: people need better ideas for what to build and how to use this powerful technology to improve go-to-market, customer service, products, and services.
That’s one reason why Zapier built and released the AutomationBench, an LLM dashboard ranking AI performance across six key domains, helping identify which models are best for which tasks.
Can you share an example of measurable ROI from a Zapier customer deployment?
Remote.com is a good example. They have thousands of global employees and an IT team of three people, which is a somewhat unheard-of level of efficiency. Through Zapier, they built an AI-powered ticketing system.
Anytime someone drops a question into Slack, an agent cross-references a knowledge base of questions that have already been answered. If a similar question has been answered before, it surfaces that same answer. If the answer cannot be found, it pings a human. When the human provides the answer, that gets encoded back into the knowledge base. The result is that humans do not need to answer the same types of IT questions repeatedly, and Remote can deploy more time and attention to serving customers better.
What does it mean for Zapier to become an “AI-first” organization internally?
In 2023, after the launch of GPT-4, we called code red internally because we believed it would be revolutionary for the industry. We needed to rethink our product roadmap so customers could benefit from those capabilities, and we also needed to rethink our own internal operations.
The first effort was getting people exposed to and comfortable with the technology. We ran a hackathon and asked everyone to pause what they were doing and work with AI. At first, people were using it like a chatbot and figuring out what kinds of questions it could answer. After the first week, we went from about 10% of employees using AI daily to north of 50%. Over the next year, through more hackathons and show-and-tell sessions, nearly 100% of people were using AI regularly.
How is Zapier approaching the shift from individual AI use to organization-wide transformation?
We are now trying to go beyond individual use cases and use AI to solve real business-critical systems. There are absolutely 10x employees who have been accelerated massively by AI. But the harder question is: who are the 10x companies?
To really benefit from AI, you have to rethink how the organization works from the ground up. It is not enough for individuals to do their jobs better. You have to address constraints in the whole process. In engineering, AI can generate code very quickly, but then you need to review the code, handle security, and decide what is ready to ship. If those other parts of the system move at the same speed, you are not getting a 10x company. That is why we are focused on moving from individual AI to institutional AI.
What are Zapier’s main priorities over the next 12 months?
One priority is helping with the “what should I automate?” problem. Where do I start? That is crucial for unlocking a new generation of builders and automators.
Another priority is governance. How does agent permissioning work? Does an agent inherit the permissions of the human who granted it access, or should it have different permissions? Agents are powerful because they can do things at a scale, speed, and cost that humans cannot. That is both a blessing and a curse. If you configure them incorrectly, they can create an immense amount of messiness. If you do it correctly, you get enormous competitive advantages.
