Why You Have to Construct Your Product for AI Brokers First

Expensive subscribers,

As we speak, I need to discuss why many merchandise should be reinvented for use by AI brokers first, people second. This isn’t a simple put up for me to jot down as a result of:

I’ve spent over a decade obsessing over constructing nice person interfaces. However the “person” in UX is quick changing into an agent as an alternative.

These pixel-perfect buttons, intuitive nav bars, and punctiliously positioned CTAs that I nonetheless obsess over are about to matter loads much less. Let me clarify by overlaying:

1. Why construct for brokers first

2. APIs, expertise, and MCPs: A fast primer

3. 5 steps to construct for AI brokers first

4. Why human judgment and oversight nonetheless matter

5. What you are able to do this week

I’m proud to companion with Granola

Your conferences include a few of the most useful context at your organization, and Granola’s MCP lets any AI agent entry it.

Merely join the MCP to Claude, ChatGPT, or Claude Code and your conversations turn into out there wherever you are working. Use it to scaffold options from what was agreed in a gathering, create tickets from standups, or draft product proposals from buyer discovery calls.

Strive Granola for Free

The reality is that utilizing Claude Code and OpenClaw has rewired my mind.

As a substitute of going to Mercury to test my funds, Linear to handle my to-dos, and Granola to get my assembly summaries, I now simply textual content my OpenClaw AI agent to get these updates. I’ll say stuff like:

1. “Embody my Mercury funds in your month-to-month electronic mail.”

2. “Record my Linear tickets in your morning briefing.”

3. “What are the highest takeaways from my conferences at the moment?”

I see an identical sample enjoying out at work:

• Your technical docs? Persons are feeding them to AI and asking questions.

• Your 30-page slide deck? Persons are studying the AI abstract as an alternative.

• Your stunning UX? An agent will skip it and work together together with your APIs.

Peter Steinberger (OpenClaw’s founder) is a good instance of agent-first considering. Some AI influencers checked out his GitHub historical past and claimed that he constructed over 40 failed tasks earlier than succeeding with OpenClaw:

Flawed. He was truly constructing 40 agent-first command line instruments that collectively make OpenClaw doable. Peter had the conviction to construct for brokers first earlier than virtually anybody else — and that conviction is an enormous cause why OpenClaw took off.

The most effective merchandise transferring ahead would be the best for brokers to make use of.

Earlier than we focus on how you can make your product agent-friendly, let’s cowl the distinction between APIs, expertise, and MCPs.

1. APIs are the kitchen instruments — particular endpoints that permit brokers work together together with your product. Slack’s API, for instance, enables you to ship messages, record channels, and handle customers. With out APIs, there’s nothing for brokers to work with.

2. Expertise are the recipes — textual content recordsdata loaded into the AI’s context that inform it precisely what to do and the way. For instance: “When the person asks to put up in Slack, name the Slack API’s “chat.postMessage” endpoint with the channel and message textual content.”

3. MCPs are the totally geared up kitchen — they bundle the API connection, authentication, and power definitions right into a single server. Slack’s MCP server, for instance, exposes instruments like “send_message(channel, textual content)” and “list_channels()” so the AI is aware of precisely what it could possibly name and the way.

Right here’s how to consider expertise vs. MCPs in apply:

1. Use expertise while you need exact management over your AI’s workflow. Expertise maintain your context window lean since they’re simply plain textual content directions. However you (or your AI coding agent) want to jot down the auth and API name logic your self.

2. Use MCPs while you need plug-and-play comfort. A pre-built MCP server arms your AI ready-to-use instruments and handles authentication behind the scenes. However connecting too many MCP servers can blow up your context window.

The most effective setup combines each. An MCP offers your agent entry to a service’s instruments. A ability teaches it the easiest way to make use of these instruments to your particular workflows.

MCPs are what your AI can do. Expertise are the way it ought to do it.

I needed to outline APIs, expertise, and MCPs first as a result of many corporations suppose “agent-first” means constructing an MCP server and calling it a day.

Listed here are the 5 steps that truly matter, so as of precedence:

1. Don’t pressure folks to make use of your web site or app. That is the mindset shift that has to return first. We’re transferring to a world the place folks work together together with your product primarily by ChatGPT, Claude, or one other AI agent. In case your product solely works when a human visits your web site or app, you’re already behind.

2. Each product functionality wants a corresponding API. Decide any vital motion in your product. If an agent can’t do the identical motion by an API, then you definately’re not agent-first. I’ve been shocked by what number of merchandise have stunning UIs sitting on prime of incomplete or undocumented APIs.