AX: Agent Experience

AI autonomy:

co-pilot -> AI agent -> AI assistance ->

Build vs. buy has changed — things that weren't viable before now are. Even whole HR systems.

The SaaS budget is replaced with a dev budget.

• openclaw.ai — autonomous coding agents, still early.
• Repurpose an old laptop as the host for it.

Autonomous cars make us rethink cities; autonomous agents make us rethink programming and systems.

Software is now read/write for everyone.

Chapter II: AX (Agent Experience)

Build tools for people that build tools.

• UX — differentiates products from competitors
• DX — differentiates platforms
• AX — differentiates platforms and products

What experience do agents have with tools and products?

So… how do we do it?

Not a feature or a protocol, and definitely not just MCP.

AX in practice:

• Access — can they access the product?
• Context — does it know about it and understand?
• Tools — work with it?
• Orchestration — can agents do work with the product?

Pillar: Access

netlify.ai — for humans and agents.

• Launch partners — payments, things to do with the thing.
• Netlify database — available to agents as batteries included.

Emerging standards — workos — self-authenticating UUID for AI agents.

Pillar: Context

Optimize for agent optimization.

Prefer plain text? Markdown? Negotiations about the content type.

MCP is UI for LLMs — context is the most important piece; expose what the agent can do now. Steering context.

API vs. MCP — same product, different surfaces.

Just shifting from API to MCP will overwhelm context — limit to a handful of tools.

Pillar: Tools

Agents will still use your product even if you don't provide tools.

Every product ALREADY has an agent experience — is it GOOD or BAD?

A CLI is often a great DX, but a bad AX — like, interactive stuff sucks for AX.

Churn churn churn.

Provide prompt escape hatches for AX.

Pillar: Orchestration

Linear, Notion — kick off longer-running tasks.

Chat bot — no one wants to use it; but orchestration lets people use the tools they are already familiar and comfortable with.

Agent runners — run in sandbox.

Actually do the thing — async agents!!

Make the whole team BUILDERS — our teammates across the business need to be able to build.

More abstract, and more required.

Evals — measuring agent experience

axis.run — actually measure the AX — try to onboard.