Interviews

Jan Čurn, Co-Founder and CEO of Apify – Interview Series

Antoine Tardif
Add Unite.AI to your preferred sources on Google
Jan Čurn, Co-Founder and CEO of Apify
Jan Čurn, Co-Founder and CEO of Apify.
Key PointsAgents will rent existing tools when building from scratch costs more than buying.
View more
  • Agents will rent existing tools when building from scratch costs more than buying.
  • Silent failures let agents confidently summarize empty pages without noticing the problem.
  • Apify paid a few hundred creators about $2M in total net earnings last month.

Jan Čurn is Co-Founder and CEO of Apify, which he launched with Jakub Balada in 2015 through the Y Combinator Fellowship. A practical need to collect apartment listings in Prague helped inspire the original product: a tool for turning websites into APIs. He has since guided its evolution from web crawling infrastructure into a platform for web data, browser automation, and a marketplace of tools increasingly used by AI agents.

Apify provides a platform for web scraping, browser automation, and ready-to-run tools called Actors. Developers can publish and monetize Actors through its marketplace, while businesses and AI builders use them to retrieve structured web data and automate tasks. In this interview, Čurn discusses how agents discover and use specialized tools, why fresh data matters, and the reliability, permissions, and economics needed for agents to work on the web at scale.

You co-founded Apify with Jakub Balada in 2015 through the Y Combinator Fellowship, initially focused on making it easier for developers to build flexible and scalable web crawlers. What inspired the idea, and how has your original vision evolved over the past decade?

The idea for Apify came from a problem Jakub and I ran into ourselves. We were working on a client project that needed listings of new apartments in Prague, and the existing tools struggled with dynamic JavaScript and irregular website structures. So we built our own tool to turn any website into an API.

Over time, it grew beyond crawling into a platform for web scraping, browser automation, and web data. We then opened a marketplace where developers can publish their tools called Actors and get paid when others use them. Now a growing share of the users of these Actors are AI agents.

Apify began as a web crawling platform long before AI agents became a major industry focus. At what point did you realize that the infrastructure you had built could become particularly valuable for AI agents that need to access and interact with the web?

The first “aha” moment came in early 2023, soon after ChatGPT launched. Suddenly you could feed fresh web data into a prompt instead of training a new model. Apify was among the first scraping platforms to provide crawlers for emerging RAG pipelines and integrate with new libraries like LangChain or LlamaIndex.

Then came the AI agent wave. Say you ask an agent to compare the price of a product across 20 e-commerce sites. Where does it get the data? That’s the problem we’d been solving for developers for years. So when Anthropic released the Model Context Protocol (MCP) at the end of 2024, we adopted it right away. We already had thousands of tools with a clear input and output, which is exactly what an agent needs.

The real turning point came at the end of 2025, when the models got good enough for agents to do meaningful work. Agents need proper tools and reliable data, and since then Apify usage has really taken off, quickly approaching $100M gross ARR.

Apify has grown into a platform with more than 90,000 tools spanning web data, scraping, automation, and AI. How do you think about the role of this ecosystem as agents increasingly need specialized tools rather than relying solely on the capabilities of a large language model?

Yes, the models will be capable of generating ever more complex software from scratch, but there is a threshold above which it’s more efficient for an agent to buy an existing tool rather than build it from scratch. That threshold will move up as the models improve, but there will always be some, giving room to sell pre-built tools, especially if they are based on private data and hard-to-get context.

As the cost of writing software goes down and the number of software systems and tools grows, the main constraint is becoming the discovery: where to find trusted and valuable tools to help you complete the job. And rather than paying $500 to generate it properly, you can pay someone $5 to rent an existing service.

From what you are seeing across Apify’s developer community, what are some of the most interesting real-world applications being built with AI agents today, and which use cases are actually making it into production?

Some of the most interesting examples are oddly specific. We’ve seen AI scan YouTube, eBay, and Japanese shops to help decide which Pokémon cards are worth buying. An insurer used Apify to check 55 Italian obituary sites to verify policyholder deaths. On the other end of the spectrum, Canary uses Apify for Spotlight, a tool that helps investigators find victims of child trafficking.

What these applications have in common is a specific problem and access to the right data, rather than an agent that’s designed to do a bit of everything.

One of the biggest challenges for AI agents is accessing fresh, structured information from websites rather than relying entirely on static training data. How important will real-time web access become as agents take on increasingly complex tasks?

An LLM is a lossy compression of the data it was trained on. It can make very clever guesses, but it’s still guessing. Once an agent is expected to make decisions or complete tasks in the real world, it can’t rely entirely on training data. Prices change, products disappear, competitors launch things, and websites update constantly.

The model provides the reasoning, but many agents need a reliable way to get current web information and turn it into structured data they can use at scale.

Web environments can be unpredictable, with changing interfaces, dynamic content, authentication requirements, and anti-automation systems. What are the biggest technical challenges developers face when trying to build agents that can reliably operate across the open web?

Modern websites can return dynamic JavaScript, challenge pages, or incomplete content, and anti-bot systems look at everything from IP reputation to browser fingerprints. Failures are often silent, so an agent can confidently summarize an empty page without noticing.

Building something that works once is relatively easy. Building something that keeps working across changing websites and at production scale is much harder.

As agents gain the ability not only to retrieve information but also to take actions online, what needs to improve in areas such as reliability, permissions, verification, and human oversight before businesses can trust them with more autonomous workflows?

Permissions become much more important once an agent can act rather than just read. The agent should never see your credentials and all secrets must live outside of a secure sandbox, never exposed to the model. We’ve been working on this proxy sandboxing problem on Actors on the Apify platform, as well as in mcpc, our open-source CLI client for MCP.

Developers are experimenting with everything from browser automation to specialized APIs and emerging agent protocols. How do you see the architecture of AI agents evolving, and what will the ideal relationship between models, tools, data sources, and browsers look like?

I think the big agent platforms like ChatGPT and Claude are becoming the new operating system, a primary interface through which people command computers, and tools plug into them the way apps do on your phone. For tools running somewhere else, MCP became accepted as the standard interaction protocol.

When an agent needs data like restaurant ratings and contacts, it can open Google Maps in a browser and click through it like a person, which is imprecise and expensive. Or it can find and call a tool that returns structured mapping data, which is faster and more reliable. The main agent will also hand off tasks like booking tables or finding flights to specialized sub-agents running on cheaper models.

Apify has spent years building infrastructure for developers while also creating a marketplace where third-party developers can publish their own tools. Could agentic AI create a much larger economy around specialized tools and services that agents discover and use autonomously?

Yes, and our marketplace is already a small version of it. Developers publish Actors, people use them, the developers get paid, and that brings more developers and more tools. That flywheel has been running for years with people on the buying side.

As AI agents take on more responsibilities, they’ll need money to complete some tasks. You can imagine giving an agent a task and a small wallet, and letting it pay for the tools it needs along the way, using an agentic payment protocol like x402 or MPP. The volume is still small, but I believe most economic activity will eventually happen autonomously between agents, with ever lower human oversight.

As Apify enters its second decade and AI agents become a much larger part of the software landscape, where do you see the company evolving over the next several years, and what do you think today’s agent ecosystem is still missing?

Even more people are now connecting Apify into their agents like Claude or ChatGPT, realizing these tools help them get more complex work on the web done quicker. The Apify platform and Actors are optimizing for a great agentic experience, driving more usage and spend.

On the other side of the marketplace, a growing number of software creators are earning regular income from their Actors: over 300 community Actors generate more than $1,000 per month. Just last month we paid a few hundred creators about $2M in total net earnings.

The growth of the Apify marketplace is accelerating, and so is our ambition to make Apify the world’s largest and most trusted marketplace of AI tools, enabling people and their billions of future agents to get work done on the web.

Thank you for the great interview, readers who wish to learn more about web data, browser automation, and tools for AI agents should visit Apify.

Antoine is a visionary leader and founding partner of Unite.AI, driven by an unwavering passion for shaping and promoting the future of AI and robotics. A serial entrepreneur, he believes that AI will be as disruptive to society as electricity, and is often caught raving about the potential of disruptive technologies and AGI.

As a futurist, he is dedicated to exploring how these innovations will shape our world. In addition, he is the founder of Securities.io, a platform focused on investing in cutting-edge technologies that are redefining the future and reshaping entire sectors.