In his closing keynote at AI Native DevCon 2025, Netlify CEO Mathias Biilmann introduced a new frontier for product design and development: Agent Experience (AX) – the idea of optimizing platforms for AI agents as users. For decades, User Experience (UX) and later Developer Experience (DX) have shaped how we build software. Biilmann argues it’s time to extend this thinking to autonomous AI. As he notes, we’re entering an era where agents interact with our products on behalf of humans, so we must consider how to craft our experiences specifically for AI agents. He defines AX as “the holistic experience AI agents will have as the user of a product or platform” – in other words, designing our tools so that a bot can navigate and use them as effectively as a person. Agent Experience (AX) Agent Experience (AX) User Experience (UX) Developer Experience (DX) “the holistic experience AI agents will have as the user of a product or platform” Why is this necessary? AI agents are fast becoming a new class of “user” in many apps, from coding assistants to customer support bots. Biilmann emphasizes that these agents shouldn’t be hamstrung by navigating human interfaces like UIs or vague docs. Instead, products need to be agent-friendly by design. He envisions humans working side by side with AI helpers in everything from coding to content creation. “I think that humans will collaborate with a ton of agents that will do all sorts of things for them – both when they are directly asked to do it, and in the background,” Biilmann said. “And as builders, [we] will have to really take a holistic view of how agents are able to do this.” In other words, teams must broaden their perspective: not just designing for the end-user’s or developer’s experience, but also for the agent’s experience operating our software. hamstrung by navigating human interfaces agent-friendly “I think that humans will collaborate with a ton of agents that will do all sorts of things for them – both when they are directly asked to do it, and in the background,” “And as builders, [we] will have to really take a holistic view of how agents are able to do this.” Designing for AI Agents vs. Humans What does it mean to optimize a product for an AI “user”? Biilmann’s key point is that AI agents experience our platforms very differently than humans do. An agent (often powered by a large language model) interacts purely through APIs, code, and data – without the intuition or context that human users and developers bring. For example, a human developer might read the documentation or notice a setup step in a UI, but an autonomous agent will only do exactly what it’s instructed or what it can infer from machine-readable cues. If a required step isn’t explicitly exposed, the agent may completely miss it. As Biilmann puts it, a workflow designed through the familiar DX lens might not translate to a smooth AX. Many platforms today still assume a human in the loop to fill gaps; those assumptions break down when an agent is at the controls. The interface of Bolt.new, an AI-powered web app builder. Biilmann used this tool to prompt an AI agent to add an email signup form to his blog. The agent dutifully generated the form code (as shown above), but the deployed form didn’t work because a hidden configuration step wasn’t handled – a step a human developer would have caught by reading the docs. The interface of Bolt.new, an AI-powered web app builder. Biilmann used this tool to prompt an AI agent to add an email signup form to his blog. The agent dutifully generated the form code (as shown above), but the deployed form didn’t work because a hidden configuration step wasn’t handled – a step a human developer would have caught by reading the docs. Biilmann learned this firsthand through a small experiment. He used an AI coding tool called Bolt.new to add a subscription form to his blog by simply asking the agent to do it. Bolt’s agent generated the code and integrated a Netlify form, which should have been great – except when he deployed the site, the form didn’t actually function. The missing piece? The agent hadn’t enabled Netlify’s form detection setting, a configuration step that was documented for human developers but never communicated to the AI. The bot had no notion of that requirement, since it wasn’t in the code or API responses. This anecdote underscored the gap between DX and AX: the feature was built for a developer who would read instructions, not for an agent that wouldn’t. Bolt.new DX AX The lesson from this story is that designing for AX often means surfacing information and flows in a way an agent can consume. It’s not enough for a product to be usable eventually (after a human reads a guide or clicks a button); it needs to be usable immediately by an autonomous script or agent. As Biilmann noted, “there’s a clear difference between approaching it through the lens of DX and the lens of AX… The more you go down this path, the more you start seeing where agents have different strengths and weaknesses”, requiring us to adjust our designs. An AI agent might excel at calling a well-documented API or parsing structured JSON, but it won’t intuitively navigate a JavaScript-laden web form or lengthy PDF manual. In practice, embracing AX means asking questions like: Are our APIs clean and well-documented for machines? Do we provide machine-readable docs or metadata that an LLM can use? By addressing these needs, we make it easier for agents to succeed – which ultimately benefits the human end-users they serve. usable eventually usable immediately “there’s a clear difference between approaching it through the lens of DX and the lens of AX… The more you go down this path, the more you start seeing where agents have different strengths and weaknesses” Are our APIs clean and well-documented for machines? Do we provide machine-readable docs or metadata that an LLM can use? AX in Action: Netlify’s Experiments Biilmann and his team at Netlify have already started applying AX principles to their platform. A vivid example is how Netlify integrated with OpenAI’s ChatGPT. When OpenAI opened up its plugin ecosystem, Netlify built a ChatGPT plugin that enables the AI to create and deploy websites on Netlify with minimal human intervention. Through this integration, a user can tell ChatGPT to, say, “deploy my project to Netlify,” and the agent will handle the steps to spin up a live site. The end user then simply clicks a link to claim the deployed site into their account – a one-click handoff from agent to human ownership. This frictionless onboarding flow means an AI assistant can take a project from code to a live URL without the user manually configuring git repos, build settings, or API keys. “We wanted to provide this frictionless ability to tell ChatGPT to just deploy this stuff you made on Netlify, so you’re not sent through a bunch of hoops, logging in back and forth,” Biilmann explained of the motivation behind the feature. In other words, the agent is treated as a first-class actor in the deployment process, not an awkward script trying to click buttons like a human. frictionless onboarding “We wanted to provide this frictionless ability to tell ChatGPT to just deploy this stuff you made on Netlify, so you’re not sent through a bunch of hoops, logging in back and forth,” The results have been impressive. Biilmann revealed that at one point, over 1,000 new websites were being created on Netlify every single day directly via ChatGPT’s integration. And that number only kept growing – recently he noted roughly 10,000 AI-generated sites launching on Netlify per day as AI coding tools proliferate. These aren’t just fun stats; they underscore that a huge wave of usage is coming from non-human agents. By making their platform welcoming to agents, Netlify tapped into a new growth vector. Biilmann attributes this success to consciously focusing on what an agent would need from the platform, identifying where the typical developer flow had to be adjusted for autonomy. For example, they ensured there were clean API endpoints for every action (so an AI didn’t need to resort to hacks), and even generated “machine-ready” documentation for the AI to follow internally. Traditional UX/DX thinking might overlook such details, but AX thinking puts them front and center. The payoff is a platform that both humans and agents can use seamlessly – and an uptick in usage from AI-driven activity that most competitors are missing out on. 1,000 new websites every single day 10,000 AI-generated sites per day both Implications for Developers and Product Teams If AI agents are becoming “users” of our software, what does that mean for the people building products? For one, it suggests that every company might need to treat agents as a new persona in their design and engineering process. Biilmann points out that many companies have rushed to bolt on superficial AI features, but the real competitive breakthrough will come from designing products that your customers’ favorite agents can plug into easily. That requires thinking deeply about the agent’s journey through your product, much like UX asks us to think about a human’s journey. In practice, this could mean providing SDKs or sandbox environments for agents, standardizing responses in structured formats, and removing human-only roadblocks in workflows. Biilmann’s advice to product teams is clear: “start consciously designing the AX of [your] products, or risk being replaced” by tools that do. Just as good DX became a differentiator for developer-focused platforms , good AX may determine which services AI agents prefer to work with (and thus which services end up delivering more value to users). agents as a new persona your customers’ favorite agents “start consciously designing the AX of [your] products, or risk being replaced” There’s also a broader industry shift at play. As AI lowers the barrier to coding and automates more tasks, it could massively grow the ranks of developers – human or otherwise. Biilmann predicts we’re on the cusp of “an era with exponentially more developers,” as AI copilots and agents enable many more people to build software than ever before. Some have even noted that platforms like GitHub now host over 100 million developers, a number poised to explode when you count AI-assisted creators. In this future, designing for AX isn’t just about accommodating robots – it’s about empowering the next wave of human developers who will be leveraging those robots. Numerous startups have already begun embracing AX as a discipline, treating it as seriously as UX. For instance, companies like Clerk (authentication) and Neon (databases) are actively dogfooding their products with AI agents to ensure they’re easy for bots to use, even hiring AI specialists to improve AX. And it’s not only startups; even Salesforce’s CXO recently described AI agents as “a new kind of user” and highlighted agent experience design as a “rising category of design” for creating great customer experiences. The idea of AX is clearly gaining traction across the industry. “an era with exponentially more developers,” dogfooding “a new kind of user” agent experience design “rising category of design” Ultimately, Biilmann sees AX as essential to building an AI-native developer ecosystem that remains open and innovative. If we design our platforms so that any AI agent (not just our own) can interact with them on behalf of users, we allow a rich open ecosystem of tools to flourish – much like the open web did. Biilmann has warned that we must ensure AI works with developers, not around them, as a key to keeping the web vibrant and free of walled gardens. The rise of AI agents is a paradigm shift, but it doesn’t have to be chaotic. By treating agent interactions as a first-class design concern, product builders can harness these new “digital coworkers” to dramatically improve user outcomes. The takeaway from Biilmann’s keynote is an encouraging one: just as UX and DX transformed software in past decades, Agent Experience could define the next era of innovation – if we’re ready to commit to it. AI-native with Agent Experience

Bolt

OpenAI

Netlify CEO Pitches ‘Agent Experience’ as Next Frontier in Product Design

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

Untitled Story

7 Major Learnings from The AI Engineering SF World Fair 2025

Agentic AI Are Cybersecurity Nightmare You Can't Ignore

Carrying Your Personal Memory Across AI Models: A TPM’s Perspective on Persistent Context in a Mult

How I Built an AI Prompt That Writes LinkedIn Articles That Actually Get Read

The Noonification: How Often Do NFTs Pass The Howey Test? (1/13/2023)

7 Major Learnings from The AI Engineering SF World Fair 2025

Agentic AI Are Cybersecurity Nightmare You Can't Ignore

Carrying Your Personal Memory Across AI Models: A TPM’s Perspective on Persistent Context in a Mult

How I Built an AI Prompt That Writes LinkedIn Articles That Actually Get Read

The Noonification: How Often Do NFTs Pass The Howey Test? (1/13/2023)

Light-Mode

Classic

Newspaper

Dark-Mode

Neon Noir

Minty

HN StartUps