Mao Hua has been building AI products since before anyone called them that. At Tencent, he spent eight years launching everything from QQ IoT (a platform that connected hardware — cameras, dashcams, air conditioners, children’s shoes — to QQ’s messaging system) to the company’s voice AI platform (powering smart speakers for Bose, B&O, Harman Kardon, and Sony) to computer vision for factory floor defect detection.
His summary of that era’s AI is blunt: “We called them smart speakers, but they were actually stupid speakers. You could ask for a Jay Chou song. You could ask about the weather. But a conversation? Impossible.”
When ChatGPT launched in November 2022, Mao Hua recognized something immediately: the conversational AI he’d been trying to build at Tencent for years was suddenly possible. He pulled core members from his old Tencent team and started building.
The Pivot That Tells You Everything
The first product was a personal assistant — managing schedules, finances, knowledge. It worked well enough. Then OpenAI released GPTs in November 2023, and Mao Hua made a decision that reveals how he thinks: he killed the product.
His reasoning was simple: anything a general-purpose AI assistant can do will eventually be eaten by the platform players. Instead, he extracted the one function that was both high-frequency and hard for platforms to replicate: financial bookkeeping for small businesses.
This wasn’t a technology pivot — it was a market architecture decision. Every company, no matter how small, must do accounting. It’s mandated by law. The work is tedious, error-prone, and chronically underserved for small businesses. And unlike personal productivity, the output is auditable: the numbers either balance or they don’t.
An AI Finance Company That Employs Accountants
Here’s where ccMonet gets interesting. They tried selling the software directly to small business owners. The response was consistent: “This looks great, but I bought your product and now I have to use it myself? I don’t have time for that. Do I need to hire someone to operate your software?”
So Mao Hua flipped the model. ccMonet doesn’t sell software anymore — it sells outcomes. You hand them your receipts (photos, email forwards, bank statements), and they give you back balanced books, tax filings, and real-time operational dashboards. The team is AI plus human accountants, and the client never touches the tool.
The results are specific: a restaurant chain with 30+ locations switched from a traditional accounting firm and immediately discovered errors in their previous books. A Thai company with 10 entities across three currencies (Thai baht, US dollars, RMB) had been unable to produce auditable reports for months — ccMonet sorted it out and helped close a 300-million-RMB funding round.
The competitive edge isn’t the AI alone — it’s the combination of AI accuracy with human oversight, delivered at a fraction of traditional cost. Mao Hua frames it as a geographic arbitrage layered on top of a technology arbitrage: Chinese engineering talent costs roughly one-third of US equivalents, and AI compresses the remaining labor further.
Why Restaurants Are the Wedge
ccMonet started with restaurants in Singapore, Malaysia, and Thailand — and is now expanding to Australia, Hong Kong, Japan, and the US. The choice of restaurants wasn’t random.
Restaurants generate an enormous volume of small, messy transactions: POS receipts, delivery platform payouts (multiple platforms per restaurant), supplier invoices for eggs, meat, cooking oil — each with different payment terms, tax treatments, and currencies. Traditional accounting firms hate this work because it’s labor-intensive and low-margin. That’s precisely why AI transforms it: the volume that makes humans miserable makes AI more accurate.
Their client list already includes recognizable brands — Heytea and Tiger Sugar among them. And the entry point is often unexpected: a venture capital firm called because a portfolio company’s financials were a mess post-investment. Mao Hua sees this as a channel — investors need portfolio visibility, and ccMonet can provide real-time financial dashboards where traditional firms deliver delayed, approximate reports.
The Cloud Is About to Get Radically Simpler
Beyond ccMonet’s immediate business, Mao Hua has a thesis about infrastructure that Western readers should pay attention to.
At Tencent Cloud, he built and sold the full stack: servers, GPUs, SDKs, APIs, databases, middleware. His view is that AI collapses most of that. The cloud of the future needs three things: storage, GPU compute, and an AI foundation layer. Everything else — the SDKs, the APIs, the middleware — gets absorbed.
His evidence is Elon Musk: “Musk just rented out 200,000 GPUs. That’s a cloud service. Is it as heavy as Google Cloud, AWS, Tencent Cloud? No. It’s just GPUs plus AI. That’s the future of cloud.”
Why Your MacBook Will Run Your Company’s AI
The second half of his thesis is about edge computing — and it’s directly relevant to enterprise AI adoption.
The biggest barrier to AI adoption for businesses isn’t capability — it’s data privacy. No restaurant chain wants to send its financial records to a cloud API. No hospital wants patient data leaving the building.
Mao Hua’s bet: within two years, Apple’s M-series chips plus open-source models will be good enough to run enterprise AI locally. He’s already doing it — he has a Mac running 24/7 at home in the US, processing personal documents with a local model.
His analogy is 4G versus 5G: for most users, the difference doesn’t matter because 4G is already good enough. When open-source models on local hardware reach “good enough” — which he estimates at DeepSeek V6 or equivalent, roughly two years out — the entire argument for cloud-based enterprise AI collapses for most use cases.
One Person, One App, One Weekend
Mao Hua has been personally vibe-coding new products on the side — building complete apps with AI assistance, from product design to code to App Store submission, without involving his engineering team. He built an expense reporting app alone and launched it at $4.90/month, using AI for everything including the App Store listing copy and ongoing conversion optimization.
His vision for ccMonet’s future is essentially an “app factory” — if the solo-development experiment works, the company can rapidly spin up specialized financial tools for different market niches, each built by a single product manager working with AI.
The organizational implication is stark. He describes it as a shift from horizontal layers (product manager, designer, frontend, backend, QA, ops) to vertical ownership — one person handling the entire stack, with AI replacing every role except backend engineering. A two-person team can ship a complete product.
“The biggest cost in any company was always human-to-human communication,” he says. “AI eliminates that. You communicate with AI, or with yourself. All that corporate language — ‘alignment,’ ‘granularity,’ ‘retrospective’ — it was all just trying to get humans to understand each other. AI doesn’t need any of that.”
This article is adapted from 离线时间 EP11.