AI Is Producing Deflation, Not Cheaper Goods: Notes from a Closed-Door WeChat Meeting
AI's cost savings are turning into a demand squeeze: layoffs and wage pressure shrink the paying user base that AI tools depend on, a problem Western builders will face regardless of market. The recommended strategy, graft onto an existing entry point and sell a measurable result, is a direct response to that slowdown.
A small closed-door meeting at WeChat headquarters made a contrarian case: AI is not simply making everything cheaper, it is manufacturing deflation. The logic is that companies cut marginal costs by substituting compute for people, layoffs and wage pressure reduce household income, supply overshoots while demand weakens, and even price cuts stop working because people can no longer afford to buy. The meeting compared the current AI wave to the 2000 dot-com bubble, arguing that sustainable paying demand has not caught up with capital flows and compute investment, and that a bust is inevitable.
The meeting's second theme was where value actually collects in AI products. Entry points, the efficiency workstations like Codex and WorkBuddy, belong to large companies. Small teams are advised to graft their products onto existing entry points. Two data points backed this: the short-drama app Hongguo became profitable by using AI to cut content cost and feed emotional consumption demand, and WorkBuddy's top use scenarios are stock trading, essay writing, and self-media, all tasks with measurable outcomes, not casual chat.
The third theme was the personal asset that survives AI: private domain and IP. With content production near zero cost, trust and personality become scarce. Community building is a product design problem, not an operations problem, and requires thresholds and tiered trust. The meeting also gave a counterintuitive pricing note: in the IP economy, raising prices filters for loyalty rather than price-sensitive buyers, and warned against putting community QR codes in crawlable public places.
The deflation argument treats income distribution as the binding constraint on AI adoption, so productivity gains that flow to capital alone end up destroying the demand that justifies the investment.
The claim that nobody uses AI to chat suggests that the conversational-assistant framing common in Western products may be off the mark; users want task completion with a measurable outcome.
Advising small teams to avoid To C entirely because it is 'big company territory' is a conservative take that may be true in China's platform economy, but it understates the niche opportunities available to focused builders elsewhere.
The IP pricing logic, where raising prices increases loyalty, generalizes to any market where identity and trust, not raw features, are the differentiator.