Last month, Cloudinary hit the milestone of four million developers signing up for its image and video platform, double the number just two years ago, and one million up from nine months ago. But perhaps the most notable trend within the past year's growth has been the new role that AI agents have been playing in driving adoption.

Traditionally, most developers have found Cloudinary through word of mouth, organic search and ads, says Sanjay Sarathy, its SVP of Self-Service and Developer Experience. The image and video platform, which also includes digital asset management (DAM) capabilities, is free to use up to a certain level of activity, with the ability to scale up to enterprise use cases as a paid account. But a growing number now find it because their chosen LLM or AI tool recommends it for managing visual content. He says:

Starting in 2025, we, along with many other companies, started to see a rise in how LLMs and AI-based IDEs were becoming sources of, essentially, registrations for us. So much so that, if you look at the last couple of months, we I think did surveys with over 100,000 of our existing base, and we asked, 'How did you first hear about Cloudinary?' Forty percent now say they first heard about Cloudinary through Claude, Cursor, name your AI tool of choice, or LLM of choice. And then after that is SEO, word of mouth, etc. So the tables have turned completely in terms of how people hear about us.

The agents effectively do the research to discover and recommend Cloudinary that a developer would previously have had to do themself. He comments:

I think one of the big challenges for a lot of software companies that have deep product capabilities is, how do you get new developers and new users to understand the range of use cases that might be possible? In some ways, our belief statement is that the rise of AI can hasten that, because the agent will tell them, 'Oh you're able to do XYZ and XYZ, versus them going off to the doc site and trying to search for something. So I think it's aided significantly in discoverability. We believe it'll aid in adoption.

Leaning into this trend, the vendor last month made it possible for agents to autonomously sign up for a free Cloudinary account and use it, with some guardrails to prevent abuse, for up to 24 hours without any human interaction. The sandbox within which the agent has operated then has to be claimed by the developer it's acting for to become a live account, otherwise it disappears. He explains:

Maybe that agent says, 'Okay, I'm going to sign up for Cloudinary autonomously, because I know it has a set of APIs that allow me to do it. It allows me to connect autonomously to our MCP servers. It allows me to define what sort of SDKs to use. That infrastructure has been built.

So, in addition to agents now coming back and informing humans about what's possible, what we wanted to do in parallel was to say, 'Hey, let's give agents the ability to just do this on the phone and then report back,' so that you're not just telling [the developer] about what's possible, you've actually done it...

We're already starting to see signups from that, based on them starting to use the product in ways that a human would too, but far faster because they're able to do it within [seconds]. There's no, 'Oh I got interrupted by this meeting.' There's none of that. They just get going right away.

As Cloudinary has always had an API-first architecture, it was already amenable to discovery by agents. But the vendor also did a lot of work to further improve its documentation, build out MCP servers and define consumable skills so that its capabilities are as transparent as possible for agents. That work has helped accelerate the adoption trend, suggests Sarathy:

I think it's impacted discoverability. It's impacted how people adopt use cases, and we're learning as we go just how many ways in which humans and agents together are starting to figure out how to build together.

The overall strategy is to make it as easy and rapid as possible for potential users to discover what Cloudinary is capable of in the context of what they're trying to achieve, he explains:

Our viewpoint is we're living in a world where the 'a-ha' moment is going to come faster and faster, and we want to provide the ability for that a-ha moment to get back to the user in a way that the user says, 'Yep, this makes sense based on what I what I want to accomplish and what I want to do.'

Another driver of adoption has been the growing ease of image manipulation using AI automation at scale, for example when creating different versions of an image for distribution across social media platforms, or for markets with different language or cultural expectations. Designers increasingly need image and video tools to organize the expanding volume of visual content assets, and Cloudinary's API-first approach lends itself to automating these workflows. Sarathy comments:

Our API-first roots are appealing to companies that have a combination of both developers and creatives in their audience. Our ability to allow creatives to manage, tag, assign metadata associated with those assets, and manage all of those assets, and as part of their ongoing operations, is highly relevant in a UI-style environment. But underneath the hood, we have APIs for all of those capabilities, if a developer needs to extend or wants to extend those capabilities to make transformations at scale...

If you're just doing it for a few assets, you can certainly, [but] our power really comes shining through when we're talking about tens or hundreds of thousands of assets that you're managing, or millions of assets that you're managing — and then transforming them, managing them, having that single source of truth for developers and non-developers at the same time.

The vendor has also introduced native functionality for API-based access to image generation models and a tool for creating video clips from still images. Other more sophisticated agent capabilities provide taxonomy and governance, search, moderation and workflow building for enterprise use cases.

While we hear a lot about the supposed SaaSpocalypse, real-world evidence seems to contradict the thesis that AI agents are a threat to established software vendors. Certainly the evidence from Cloudinary's story suggests that agents are fueling strong adoption of its platform — and even signing up for it directly, although still requiring the subsequent approval of the developers they act on behalf of. It's a useful reminder that the best way to survive tech innovation is to be ready to adapt.

One of the astute ways in which Cloudinary has done this has been to view agents as a class of prospect in themselves, rather than seeing them as simply equivalent to the developers and creators they already serve. As I've noted elsewhere in the context of agent management, agents are neither direct replacements for people, nor just another class of IT asset. Their probabilistic approach means that they need to be given more comprehensive information and context than traditional, deterministic IT assets. This in turns means that they need to be given rights and permissions that may seem equivalent to those of a human user. But because they have less innate awareness of what boundaries should not be crossed, the guardrails and approvals put around them need to be more stringent than you would normally put in place for people. By pitching its offer at just the right level for agents — providing carefully curated information and context, but strong enough constraints to remain accountable to the person in charge — Cloudinary has opened up a new channel that is rapidly expanding its reach.

[Amended September 11th to describe Cloudinary as an image and video platform, rather than a DAM.]