Zapier CEO says ‘very few’ companies are using AI effectively

Millions of businesses use Zapier to automate workflows, but its CEO and co-founder says not many of them have fully tapped into AI’s potential.

Wade Foster said most workers, whether at Fortune 500 companies or “mom-and-pop shops,” are mostly using ChatGPT and its ilk for one-off tasks like writing emails rather than fully automated processes. He also described what sets automated workflow processes—Zapier’s bread and butter—apart from the somewhat squishy term of AI agents.

The platform is currently focused on widening its customer base from smaller businesses to bigger enterprises.

We spoke with Foster about missed opportunities for revenue teams, what a hybrid agentic workflow is, and the future of work.

This conversation has been condensed for length and clarity.

With everyone talking about AI agents right now, is there a lot of category confusion between what an agent is and what an automated workflow is? What’s a concise way of thinking about the difference?

There is a lot of confusion about what is an agent and what isn’t an agent…There’s two ends of the spectrum. On one end, you have what I would call a classic, deterministic workflow. This is: When this happens, do this, do this, do this. It works like code, works like a program. It follows a human’s instructions to a T.

What I would think of as a purely agentic system is one where you have a goal in mind and you give an AI system that goal. You give it instructions on how to achieve that goal. You give it access to data, tools, integrations, things like that, and then it uses its own reasoning to go complete that goal.

You can achieve the same goal with either system; it just sort of depends on what things are you optimizing for, on which one you want. And it turns out what we’re seeing a lot of is what I would actually call a hybrid—what I would call an agentic workflow. And I think the big reason those are popular right now is that it improves the overall reliability of the system and reduces the overall cost.

They’ll have a node in a workflow that actually does have an agentic property, where it’s like a mini-goal with mini-instructions to go complete a particular task, and they’re using the prior steps in a workflow to feed in the data and the context to allow the agent to do it…Hybrid is very, very popular right now.

With every workplace tool implementing its own AI features, its own AI agents, does that make it more challenging to integrate all of them together?

I think the job of integration is the same as it’s always been. You’re always trying to understand what users are trying to do with all these different systems. How do you provide good interoperability and importability between these things? The only thing that’s really changed is that some of the underlying technology is a little different.

Now you have tools like model context protocol (MCP), which allows agent-to-agent communication, which is very helpful if you don’t necessarily know what the underlying protocol is…There are emerging, new methods that make integration easier in some ways.

What should people be using Zapier for?

The places where we see folks in marketing and revenue that are getting a lot of benefits from using Zapier tend to be in their lead management flows. Almost everybody is running Google ads or Meta ads or LinkedIn ads or something like that. Or they have forms on their website where they’re collecting leads…What we’re seeing the most sophisticated companies do is really enrich those workflows and bring them to life with AI.

So now what you can do is you can take that lead coming in from one of the ad platforms. You can hit an enrichment API. You can hit a waterfall enrichment API. You can go do deep research on the internet to basically pull in all this extra context about this person or this organization…And then once you’ve soaked up all this information, then you can go add that to your CRM.

There’s a lot of organizations that sell products at a price point where you just can’t provide that really nice experience that you might get when you buy a premium product from a sales rep who just really understands you deeply and intimately. You can’t do that if you’re selling, you know, $50 products. But now with AI, you can get a little closer to that.

How many sales teams, revenue teams have reached that point of sophistication?

Very few. The place where you see a lot of AI adoption is what I would call everyday AI use cases. For a lot of organizations, this means they’ve procured something like ChatGPT or Claude or Gemini or Copilot for their organization to use…But it’s all one-off tasks, personal productivity tasks, which are valuable, but to me, it doesn’t raise to what I would call this bar of transformative AI.

It’s still early days for thinking about how agents can work together. How do you see that playing out in the next year?

I think we’re going to see a pretty massive explosion of agents tackling all sorts of tasks in the workforce. We’re seeing it right now with coding agents. Coding agents have just massively exploded in the last month or two.

The challenge is, there still is a fair amount of friction, because if you want access to those capabilities, you have to [do so] through tools that are mostly made for engineers. So there’s an ergonomic problem, a user experience problem, that is a challenge. Over the next year, you’re going to see a lot of that just get smoothed over.

The second thing you’re going to see is just an explosion of work. I think we’re already seeing this with the coding agents. The engineers are writing far more code than they ever were before, because the tools are so good at doing this, and they’re having it tackle tasks that everyone knew was a problem before, but just tolerated...That’s going to create a lot of work for humans trying to spot and find these problems and go, like, “Oh, we can solve this, we can solve this, we can solve this.”