When AI Actually Starts to Work, the Problems Are Just Beginning

AI as a spotlight, not a solver

Attempts to use OpenClaw-like tools (OpenClaw类工具) in both office and personal settings have produced a blunt lesson: AI often reveals problems instead of fixing them. Users report that automation surfaces inefficiencies, ambiguous responsibilities and fragile processes that were previously hidden. The tools can execute tasks, but they need a cleaner pipeline to pull from; until you resolve the human and organizational mess, the automation sits idle.

One-person shops gain an edge

Paradoxically, solo operators benefit most. A one-person company has no cross-departmental turf wars, no permission bottlenecks and no passive handoffs — the exact frictions that bloat AI prompt engineering into an administrative task. That makes an individual far better placed to iterate on workflows and “teach” the system. But what does that scale look like in a larger company? It quickly becomes costly.

Prompting becomes project management

To get useful output, users now must provide context, goals, constraints and usage scenarios — often more clearly than they would explain the task to a colleague. Prompt-writing is morphing into a form of task design. And that creates new work: verifying provenance, asking “how much of this did you produce?” and policing quality. When everyone’s production cost approaches zero, consumption and curation costs spike. So organizations hire AI to summarize AI, and the stack deepens.

Data, lock‑in and geopolitical contours

There are also structural risks. It has been reported that many platforms work best when users put data and histories into their cloud; that increases efficiency but also creates lock‑in. For Chinese users this intersects with domestic cloud competition — led by players such as Baidu (百度), Alibaba (阿里巴巴) and Tencent (腾讯) — and a regulatory environment focused on data security and cross‑border flows. Geopolitics matters: trade controls and sanctions have already nudged Chinese firms to build autonomous stacks, and reliance on cloud agents raises questions about privacy, compliance and control. Ultimately, the real test won’t be whether AI gets stronger, but how we decide to handle the messy, human problems it exposes.