Accountant for Nearly 2 Decades Learns Vibe Coding to Improve His Work

For over 18 years, Wei Khjan Chan has worked as an accountant, a profession often reported to be at risk of automation. Every time he saw headlines warning that AI could replace jobs like his, he said he felt the pressure build.

“It would be great if I learned about AI earlier. At least I replace myself rather than letting other people replace me,” the 39-year-old told Business Insider.

To stay ahead of the curve, Chan embraced mood coding, using AI tools to write code and build apps. The audit partner at an accounting and consulting firm in Malaysia said he came across mood coding in June after attending weekend coding workshops in Singapore and Malaysia.

Although he had no technical background, Chan created a web application to solve a problem in his professional life: filing expense reports after a business trip.

The app uses AI-powered optical character recognition to scan and process receipts, automatically exporting them into files for its company’s finance teams. It also uses AI to automate its workflow, such as generating invoices.

“This code is a bunch of JavaScript, which I obviously don’t understand,” he said, showing Business Insider his web application. “Without the mood coding tools and skill set, an accountant is incapable of doing this,” he added.

AI is not a way out of accounting: it’s about how to save it

The accountant said he didn’t learn mood coding to change careers. Rather, he considers “AI know-how” to be a fundamental skill for any office profession, such as Excel.

Building his own apps has shown him how powerful the tools can be: What once took weeks and an entire team to build a proof of concept can now be prototyped in a single weekend, he said.

Chan also told Business Insider that he is advocating for wider adoption. As a board member of his local accounting institute in Malaysia, he advocates for more AI training on a large scale.

Fewer people are moving into accounting, even though the demand for accounting services is increasing. With the labor shortage, AI could help fill the gap, Chan said.

Lessons Learned from Vibrational Coding

Chan said that when he began experimenting with AI, he was advised to write long, detailed prompts with “full length of context.” But experience has taught him that smaller, iterative steps work better.

“The initial prompt is very important to get everything set correctly,” he said. After that, when changes are needed, it’s more effective to adjust a small part at a time rather than piling up an entire wish list.

He approaches it like managing an intern: dividing tasks into smaller, more precise instructions. The more specific you are, the better the result, he said.

Not all lessons have been easy. As part of a project, Chan built his database based on a single organization. When someone later requested multi-company support, they found they had to rebuild the entire structure.

“It’s a very fundamental change,” he said. “I ruined everything.”

This experience taught him that it is essential to get the architecture right from the start, because features and functions can always be layered on later.

As for debugging, it’s basically like “complaining to the AI,” Chan said with a laugh. If the error message changes, it’s usually a good sign: the AI ​​is fixing the problem. If the same error persists, he said he will reset the conversation and reframe the request with new examples.

And he said that, despite occasional debugging, ambiance coding doesn’t require endless hours of work.

Chan usually tinkers after his kids go to bed, adding a feature here or refining a function there. “It’s like playing a game,” he said.

Over time, this builds up, and with a little guidance, the pieces eventually come together.

Do you have a story to share about ambiance coding? Contact this journalist at cmlee@businessinsider.com.