#Why I Build Products This Way — From One Realtor's Pain to a Platform Built for AI

A lot of people build AI products by starting from a flashy piece of tech and then hunting for a problem to bolt it onto. I tend to do it backwards: first ask what problem you're really solving and what should be automated — then bring in the tech.

This post is an honest look at the two things I've been building lately — a platform for real-estate agents, and an AI entry point for ordinary people — and what's actually behind them.

#1. Problem first: I went and interviewed realtors

I didn't start by deciding "I'm going to build a realtor SaaS." I started by asking.

I interviewed the real-estate agents around me and ran the Hong Kong market through a round of analysis. What surfaced: their day is largely eaten by three repetitive things.

• Listing: after the photos, they write the copy themselves and post to each platform by hand — an hour per listing.
• Serving clients: after-hours and midnight inquiries go unanswered; the client moves to the next agency.
• Following up: no contact captured, no follow-up on viewings — the lead quietly vanishes.

These three share something: none of them is the deal itself, yet they eat the most time before the deal. So that's what should be automated.

That became UD House — snap one photo, AI writes the listing copy, pushes it to every platform, runs a 24/7 concierge, proactively books viewings, and captures the lead. The agent just presses the shutter; the system does the rest.

The point was never "we have AI." It was "we removed the most painful, most repetitive stretch of an agent's day."

#2. The first bet: AI agents are the future, so products must be readable by AI too

I believe AI agents are inevitable. In the future, the thing getting work done for you may not be you tapping buttons — it'll be an agent. Yours, or someone else's.

If that's true, a product can't only be for humans to look at. The website, the admin, the API — all of it has to be readable by AI and programs too.

So from day one UD House does three things at once:

• a human-facing interface (the agent's admin, the customer's share page);
• an open API + a public OpenAPI spec for agents;
• an llms.txt for crawlers / LLMs to self-onboard.

An AI agent can read our spec and then create listings, generate copy, and pull leads for an agent — without a human tapping a single button. In realtor SaaS specifically, few products expose this layer today — but I think it'll become table stakes.

#3. The second bet: but ordinary people are still far from "raw agents"

Here's a reality that's easy to miss: even though agents are the future, ordinary consumers are still a long way from raw agents (the OpenClaw kind).

You can't ask a normal person to set up an environment, write prompts, and wire APIs. What they want is: open their phone, tap a few buttons, done.

So I built Pin — a consumer-grade entry point (currently in private beta). It wraps a product's capabilities into a button interface on LINE / Telegram: cross-domain, intuitive. You don't need to understand agents; you just tap. The same capabilities underneath, dressed in something a human understands.

In other words: agents are for the future; an intuitive entry point is for the people of today. You need both.

#Closing: one spec, feeding humans and AI alike

Tie those three together and it's really one idea.

Each product declares one SKILL.md — the single source of truth for what it is (its entities, its actions). That one spec then grows four surfaces:

• Pin — the LINE / TG button app for ordinary people;
• MCP / API — for AI agents to operate;
• Web admin — for human ops;
• internal fleet — for automation.

One spec, four surfaces in sync. A person comes in through Pin, an agent comes in through the API — same road underneath. (Today the API and web admin are live; Pin and MCP are in private beta — but they all consume the same spec.)

That's how I build: first decide what to solve and what to automate; then make it readable by both humans and AI; then give today's people an intuitive entry while paving the road for tomorrow's agents.

The tech will keep changing. "Be clear about what problem you're solving first" never does.