What's this about?
Wang Puzhong, CEO of Meituan's core local commerce business, gave a public post-mortem of the company's internal AI transformation. From February to March, Meituan ran a company-wide "raise shrimp" campaign. The results he reported: the bill exploded to over 10 million RMB per day, and the errors the effort produced interfered with actual business operations. Baoyu's reaction asked the natural follow-up — does this mean OpenClaw is finished?
What exactly was the "raise shrimp" campaign?
It was an internal push to get every employee using AI agents, named after the tool at its centre. The pattern is familiar: leadership mandates company-wide adoption of a new technology, measures activity rather than outcomes, and everyone complies enthusiastically. What makes this post-mortem unusual is the two specific failures named. The cost is one — usage-based pricing means enthusiasm converts directly into spend, with no natural ceiling. The second is worse: outputs that were wrong but plausible enough to feed into real decisions, so the campaign didn't just waste money, it degraded the business it was meant to improve.
Why is this worth noticing?
- Public candour from an executive is rare here. Most enterprise AI post-mortems stay internal; naming a daily burn figure and admitting operational harm is unusual and therefore informative.
- It's the counterweight to this week's optimism. Databricks argued agents finally work in the enterprise because organizations captured the missing context; Meituan shows what the same push looks like without that work.
- The failure mode is the interesting part. Not "the AI couldn't do it" but "the AI produced errors convincing enough to act on" — which is the harder problem, and the one no dashboard catches.