“Design your interfaces for agents, not humans. The consumers of your endpoints in 12–18 months will mostly be other agents. Build well-typed contracts, not pretty UIs.” — Lou, 2026-05-21
Session context: 2026-05-21_Mastermind — Dirk asked whether AIMM members should package their offerings as MCP servers clients connect to their Claude. Lou answered with the MCP vs API vs Skill decision framework, then added the forward-looking principle: the relevant design question is not “how does a human use this?” but “how does an agent call this?”
Core Idea
Every business owner who builds client-facing interfaces today is designing for the wrong consumer. They’re optimizing for humans to read, click, and scroll. In 12–18 months, the primary consumer of those interfaces will be AI agents — their clients’ Claude, GPT, or Gemini acting on instructions, calling endpoints, interpreting outputs.
Agents don’t read landing pages. They don’t click through a UX flow. They don’t benefit from prose explanations of what a service does. They need three things: a well-typed input contract, a predictable output format, and fast response times. Everything else is decoration.
The practical implication: stop building UIs for users who will not see them, and start building well-typed interfaces for the agents that will. An API endpoint with clear schema documentation will serve an agent better than a beautifully designed dashboard. A skill with explicit input/output descriptions will be invoked correctly; one with vague instructions will be called wrong and produce garbage.
This is not a distant future consideration. Members who are building MCP servers, APIs, or skill packages right now are making interface decisions that will either age well or require painful rewrites. The ones who design for agents — clean contracts, structured outputs, predictable behavior — will find their assets composable as the agent layer matures. The ones who design for humans will find their assets inaccessible to the agents their clients eventually deploy.
The question is not “how do I make this easier for my client to use?” It’s “how do I make this callable by my client’s AI?”
Practical Application
Audit your current client-facing assets for agent-callability. For each one: does it have a defined input schema? Does its output follow a consistent, parseable structure? Can it be called programmatically without human involvement? If the answer is no, it’s not agent-ready.
Prioritize structured output over human-readable prose. For any deliverable that might eventually be consumed by an agent (a diagnostic report, a coaching summary, a framework output), add a structured JSON block alongside the prose version. Agents read the JSON; humans read the prose. Both are served.
Write your API docs for machines, not just humans. This means: explicit data types on every field, enumerated values where possible, error codes with definitions, examples of valid vs. invalid inputs. Human developers will also benefit, but agents will require this.
Design your skill descriptions as contracts. See the insight on how Insight - Tools Define AI Capability More Than Model Intelligence: if the description of your skill is vague, the agent calling it will underperform. The description is an interface, not a marketing headline.
Coaching question: “Pick your most valuable client-facing deliverable. If an AI agent tried to call it tonight, what would break?”
Evolution Across Sessions
This forward-looking principle surfaced explicitly in 2026-05-21_Mastermind as the conclusion to the MCP vs API vs Skill framework. It recurs as the implicit design philosophy behind Insight - Your Expertise Becomes an Interface — How Clients’ AI Accesses Your Intelligence Layer and Insight - MCP vs API vs Skill — Three Patterns for Exposing Your Knowledge to Clients’ AI. This insight names it as a standalone principle: the shift from human consumer to agent consumer is not a feature of a specific technology choice — it is a design axiom that should govern how knowledge entrepreneurs build any client-facing system from now on.