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.

Originally published on Ultra Lab — we build AI products that run autonomously.

Try UltraProbe free — our AI security scanner checks your website for vulnerabilities in 30 seconds: ultralab.tw/probe