AI Services Are the New Software (And Why That Matters)

A guy in Singapore DM'd a tze char stall in Geylang, sent over an AI-generated online ordering page, and walked away with $1,200 SGD. Total build time: 2 minutes.

Let that sit for a second.

Not $1,200/month. Not a recurring SaaS contract. Just one DM, one page, one PayNow transfer. And the wild part? The site was already built before the pitch even started. The AI did the fulfillment upfront. That's the entire game changing right in front of us, and most people are still trying to build the next ChatGPT wrapper.

Services Are the New Software

Sequoia just dropped a piece making this exact argument: for every $1 spent on a software tool, $6 gets spent on the service delivering the same outcome. Six to one.

If you're selling software, you're competing with the next model update. Anthropic ships something new on a Tuesday and your moat evaporates by Wednesday. But if you're selling the outcome — the website, the workflow, the booked reservations — every model update just makes your business better. Faster. Cheaper. Higher margin.

This is what investors mean when they say the most defensible AI businesses might not be the ones building the AI. They're the ones using it to actually do work for people who don't have the time, taste, or patience to do it themselves.

Think about it from the business owner's side. If you run a hawker stall in Bedok or a salon in Tampines, your day looks like:

• Managing staff and rosters
• Keeping ingredients or stock topped up
• GST submissions and bookkeeping
• Running GrabFood and Shopee promos
• Dealing with the auntie who wants her bee hoon "extra dry but not too dry, ah"

The absolute last thing you want to do at 11pm is open Claude and prompt-engineer your own website. You'd rather pay someone $1,200 to make the problem disappear. That's the business.

The Arbitrage Window Is Open (For Now)

Opportunity arbitrage is a beautiful, temporary thing. It's that gap between when a new technology changes how work gets done and when the market actually prices that change in.

Right now, that gap is huge.

What used to take a 3-person agency two weeks now takes one person two minutes. What used to require memorising the WordPress UI for three years now requires good taste and a Canva brand kit. The skill ceiling collapsed. The skill floor rose. And the people who move first capture the spread.

If you're a non-techie reading this and thinking yeah but I can't code — that's actually the point. You don't need to. Tools like Claude Code now let regular humans ship real things. I wrote about shipping to production in 5 hours as a non-techie and honestly, it's the closest thing to magic I've experienced in business.

What the Stack Actually Looks Like

Here's the loop that's making this work, broken down into the parts that matter:

The whole thing fits inside one Claude project with a few well-written skill files. If you've never used Claude this way, Claude Cowork is the layer that turns one-off prompts into actual repeatable workflows. It's the difference between asking Claude to do a task and having Claude own a task.

The Skill That Still Matters: Taste

I want to be honest about something because the demos make this look effortless and it's not.

AI gives you the website in two minutes. Cool. But the first version only put five menu items on the page. Five. The stall had over forty. AI is exceptional at confidently lying to you, and if you ship that to a real client, you've burned the relationship and probably the kopitiam-grapevine referral pipeline behind them.

So here's the actual skill stack you need:

• Taste. Knowing what a good website looks like. What converts. What feels professional vs. what feels like a 2014 Carousell listing.
• Quality control. Cross-checking AI output against the real menu, the real hours, the real photos. Every. Single. Time.
• A refined skill file. The more effort you put into your skill.md, the better your outputs. If 100 people use the same model on the same task, one person's output beats the other 99 — because that one person actually wrote the instructions properly.
• A brand kit hack. Search "brand guidelines" templates on Canva, save as PDF, feed it to Claude, tell it to design within those rules. Your sites instantly look like a professional designer made them.

The Move

If you take one thing from this post: stop trying to build the next AI tool. Start using AI tools to deliver outcomes people will pay for today.

Pick a niche you actually understand or care about — hawker stalls, salons, dental clinics, aircon servicing, property agents, tuition centres. Pick something with bad existing websites and busy owners. Build one site this week. Send it to one business. See what happens.

That's the whole game. One site. One pitch. One yes.

The window is open right now. It will not be open in 24 months. The people who move early eat. The people who wait read articles about the people who ate.

Then go build something. Today. Not Monday.