← Back to the library

GUIDEAutomation中文版

A 30-day plan to start an online business with AI agents, built from what we shipped

A four-week plan to build and test a first small product with Claude Code and Codex, based on what we shipped from 28 Sep to 15 Oct 2026, plus the Malaysian basics for month one.

WHAT YOU’LL GET

A four-week plan that starts with a real user, the split between what agents did and what stayed with the owner, and the Malaysian paperwork for month one.

WHO IT’S FOR

Owners and solo builders in Malaysia who want to ship something real with AI agents.

DIFFICULTY

Intermediate

TIME

20 minutes

WORKS WITH

Claude Codex

Get the full file

The whole resource as one Markdown file for your notes or your AI workspace.

FREE

What "30 days" means here

This is a 30-day plan, not a claim that we built a business in 30 days.

Our own timeline: the dainer.ai code history began on 30 Aug, and the site and Academy were live on 9 Sep. The Chinese use-case directory followed on 15 Oct, about 46 days after the first commit. Most of the products below shipped between 28 Sep and 15 Oct.

The build receipts don't show customer revenue. Pluto's one recorded payment was our own checkout test, marked for refund.

What we shipped, with dates

  • Harga Hari Ini, a grocery price tracker built on public KPDN PriceCatcher data. Asked for on 28 Sep and live on 29 Sep with 295 items. On 2 Oct the build generated 3,915 more pages, and a local check of 3,916 pages counted 187,262 internal links with no broken ones.
  • Pluto, a paid website scan. Its checkout went live on 30 Sep.
  • Mini Games on dainer.ai. Work started on 9 Oct and merged on 13 Oct, after 4 review rounds.
  • Kira Dulu pre-check. The engine passed 248 of 248 tests and failed its first independent review before passing the second. The interface passed 289 of 289 tests and went live on 15 Oct.
  • A Chinese AI use-case directory, with 2 live entries and 8 reviewed drafts by 15 Oct.

Traffic is a different story. From 8 Sep to 5 Oct, dainer.ai had 99 search impressions, 2 clicks and 66 sessions.

Who did what

The agents (Claude Code and Codex) wrote the code, ran the tests and fixed failures. Claude planned the work, and Claude or Codex built it. The model that didn't build a change reviewed it, and failed it when something was wrong.

I decided what to build and what to charge, and kept approval over production releases and account changes. Agents also did some approved setup and deployment work. Business registration and payment-account decisions stayed with me.

The 30-day plan

Week 1: a user before a build

  • Pick one problem you can explain in one sentence, for a buyer you can name.
  • Before building anything, ask a likely user to show you how they handle the problem now. Write down the step that causes trouble.
  • Ask Claude to plan the smallest version that completes that step for them.
  • Register the business. SSM's annual registration fee is RM30 under your own name or RM60 under a trade name (SSM fee table/table-of-fees/Table-of-Fees.aspx)). The route has eligibility rules, such as age and where the business is based, so check SSM's registration brochure first.

Week 2: build, with a second model reviewing

  • Let the agent build the small version. Before calling it done, have it run the tests and show you the real page.
  • Set up a reviewer that didn't write the code. We use a read-only Codex run that ends with PASS, PASS WITH FIXES or FAIL. The builder never approves its own work. Our Kira engine failed its first review, and that review caught real problems.
  • Put every change through a pull request, a proposed code change you can review before merging, and merge it yourself.

Week 3: the parts a business needs

  • Add a privacy page, a contact path and terms.
  • If you collect personal data, read the PDPA rules. JPDP says the data protection officer requirement took effect on 1 June 2025 and applies when certain processing criteria are met (JPDP FAQ). Check those criteria before you collect customer data.
  • Connect payments through a payment gateway, and check its current FPX and DuitNow fees yourself.
  • Check e-invoicing. LHDN's current timeline lists an exemption for taxpayers with annual turnover or revenue below RM3 million (LHDN timeline). Check the detailed criteria for your business.

Week 4: let users try it, and measure

  • Ask the user from week 1 to try the product without your help. Watch where they get stuck.
  • Write a few pages that answer what your buyer actually searches for, using what you know from the work.
  • Set up analytics and check it every week.

Use day 30 as a review point. Check whether the product works for a real user, and whether their feedback justifies more work.

What to watch out for

  • "The agent said it's done." Agents report success they haven't checked. Ask for the test output and the live URL.
  • Token cost. Our biggest token use came from an agent clicking through a browser. Once the agent knows the steps, turn them into a script.
  • Account actions. Keep approval of payments, DNS changes and production releases with the owner.

Keep checking the product with likely users while you build. Let what they try decide what you fix next.