Conversation with Wang Ming of K2 Lab (攀峰智能): Human Efficiency Ratios Are Terrifying in the AI Era — Burning US$10 Million in Tokens Is Just the Beginning

Fast burn, early traction

K2 Lab (攀峰智能) founder Wang Ming (王铭) warns that in the AI era human efficiency ratios are "terrifying" — and the company is backing the claim with real spending and rapid product execution. The AI-native startup has completed another angel round worth several tens of millions of yuan reportedly within three months of an earlier close, and it has been reported that the team plans multiple fundraising rounds this year to pursue a rapid "land-grab" strategy focused on financing efficiency rather than slow, traditional scale-up. Their first product, Moras, is positioned as a KOC (micro‑influencer) Agent OS that automates the full content‑commerce loop for overseas creators: product selection, content creation, video publishing and analytics.

Moras is built on a Harness Engineering multi‑agent approach. It codifies discrete "Agent Skills" — from market insight to hook design, automated editing and policy pre‑checks — into a scheduler that parses intent and executes tasks end‑to‑end. It has been reported that K2 Lab’s current token consumption, inferred from billing, is roughly on the order of 1 billion tokens per day, and Wang says that to meet a targeted ARR in the "tens of millions" of US dollars the company expects token consumption valued at tens of millions of dollars as it scales. Early tests reportedly showed strong unit economics: invited creators posted a first‑week order rate over 70%, with active creators averaging nearly US$10,000 monthly GMV and certain cohorts exceeding US$100,000 in a testing month.

Strategy, risks and geopolitics

Wang frames the company's play as platformic: where mobile internet platforms were centralized "super‑entrances" that aggregated attention, the AI era will be defined by distributed Agent OSes that intercept and execute user intent. K2 Lab has chosen TikTok (overseas) as the initial battleground because good commercial operators are still thin on the ground there, giving early entrants room to establish a bilateral A2A (consumer↔product) ecosystem. Internally the team leans heavily on AI — reportedly 99% of code is AI‑generated and the company even uses self‑built development agents for testing and delivery — freeing them to hire fewer traditional specialists and iterate quickly.

That aggressive token burn and overseas push come against a shifting geopolitical backdrop. US export controls on advanced chips and heightened scrutiny of cross‑border AI services have tightened access to high‑end compute for many Chinese startups; as a result, token economics, cloud partnerships and model sourcing become strategic levers as much as product features. Can heavy upfront token spending and fast user monetization outrun regulatory and supply‑chain headwinds? Wang believes the soil is richer than the mobile internet era and that winners can emerge fast — but the calculus now mixes product execution with capital efficiency and geopolitics in ways Western readers may not have fully appreciated.