Agent Experience

We need to set up our agents for success.

The term I keep coming back to is agent experience. Netlify's Mathias Biilmann put a name to it in 2025, and I think it deserves far more attention than it gets. We, as companies, organizations and individuals, should be investing a lot more time in setting up our agents for success.

Give them access, not handholding

Think about it this way. You want your child to succeed and reach their goals. So you give them access to the internet to learn and to build things. You obviously don't handhold them. You let them be curious and explore the path on their own, but you give them access to the tools and resources that help them explore and succeed.

Agents are no different. They don't need us standing over them. They need:

• Access to the right tools, systems and information.
• Context on how things actually work here.
• A clear picture of what good looks like.

Leave any one of those out and even the best model will struggle.

Design for humans, and for agents

In the age of AI, this has become extremely important. As good engineers and product leads, we spend our time on the human interface and the human experience, the places other people spend their time. That still matters. But we now have to spend our time on agent experience too.

• UX is for people using your product.
• DX is for developers building on it.
• AX is for agents acting on it.

We need to help our agents do better jobs and give them better context.

Context engineering, harness engineering, agent experience

There are related terms here, and they're worth separating:

• Context engineering is about what goes into the model's window for a given task.
• Harness engineering is about the environment around the model: its tools, permissions and checks. Every time the agent gets something wrong, you fix the environment so it doesn't happen again.
• Agent experience is the wider lens. When an agent, yours or anyone's, meets your product, your docs, your codebase or your website, how easily can it find what it needs, do the job, and recover when something breaks?

Harness engineering is the closest of these to what I mean by agent experience. So why call it agent experience? Because it's not just a technique. It's a mindset.

What it looks like in practice

Setting up our agents for success is mostly small, unglamorous work:

• An AGENTS.md that explains how the codebase really works.
• Tests and linters that define "right", so an agent can check its own work.
• Error messages that say what went wrong and what to do next.
• Docs available as plain markdown, not only as rendered pages.
• An llms.txt that tells agents what your site is and where to look.
• APIs and MCP tools with clear names and typed inputs.

I did a small version of this on my own site: an llms.txt, a sitemap and structured data. It took an afternoon. Now when someone asks an assistant to find AI engineers in Dubai, an agent reads something clean instead of guessing from a hero section.

Our mindset needs to change

This is the real reason I call it agent experience: our mindset needs to change.

For most of our careers, our taste lived in the code we typed. If we wanted something done our way, we did it ourselves.

We need to adapt and let go of our code. That doesn't mean caring less about quality. It means moving our taste out of our hands and into the environment: the conventions we write down, the examples we leave behind, the tests that fail when something's off.

We need to set our mindset in the direction of helping our agents do the job, our machines do the job, and do it better. Do it with our taste. And do it right.

We spent a decade getting good at user experience. The teams that get good at agent experience next will be the ones whose agents actually ship.