What this is

Product manager Pa Langmao published on Juejin this week, breaking down 20 years of content product evolution into three generations. His conclusion is blunt: AI has compressed single-article production from 2 hours to 30 seconds, and we are now entering an "AI production era" defined by capacity explosion and a reversed scarcity curve.

The three generations correspond to different holders of editorial "deciding power":

  • Editor's-pick era (2005–2012, Sina, NetEase) — the homepage was the resource slot, and the chief editor was king.
  • Algorithm-recommendation era (2012–2020, Toutiao, Douyin) — distribution cost was crushed to near zero, but filter bubbles came along for the ride.
  • AI production era (2023–present) — the supply-side capacity problem is fully solved; marginal cost approaches zero.

What does this mean? When anyone can produce an article in 30 seconds, the scarcity is no longer production capacity — it is "how to surface content worth reading" and "how to make users believe it is good."

Industry view

The mainstream narrative says "AI democratizes creation," but the cooler read from the product community is: after the supply explosion, the real scarcity flips to filtering and trust — not to production capacity itself.

Pa Langmao flags three new problems:

  • Quality assessment must be tiered. L1/L2 machine screening handles readability, factual accuracy, and compliance; L3 human review covers viewpoint value; L4 user feedback validates the loop. "Does it read human?" is the wrong test. "Does it deliver real value to the user?" is the right one.
  • Mixing AI and UGC (user-generated content) in feeds is a politics problem, not a tech problem. Mandatory AI labeling drops click-through rates 30–50%; no labeling collapses trust, and rebuilding it costs ten times more. Xiaohongshu and Zhihu are currently walking on this minefield.
  • Copyright gray zones. Domestic (Chinese) courts generally do not recognize purely prompt-generated content as a protected work — so enterprises using AI to write marketing copy face compliance risk.

There is pushback. Critics argue this tiered assessment is overengineered — most small and mid platforms lack the resources for L3 human review. Others note that AI content labeling is unenforceable in practice: relying on creator self-disclosure is the same as no labeling at all. The trust mechanism ultimately has to rest on platform endorsement, not rules.

Impact on regular people

For enterprise IT and marketing teams: content marketing budget structures will shift. Outsourced writing costs fall, but internal costs for review, labeling, and compliance rise. A team that once needed 3 content editors may become 1 content operator + 1 reviewer.

For individual careers: "production-type" work — copywriting, slide decks, report assembly — is significantly more replaceable, and bargaining power is compressed. The gatekeeping skills — "deciding what is worth writing" and "building personal recognition" — become more valuable instead.

For consumer markets: in the next year, AI content density in feeds will visibly rise. Our suggestion: when you see high-quality content with vague sources, verify before forwarding; for purchasing decisions, prioritize sources with real identity backing.