What this is
This week, Tencent WorkBuddy launched its 'Expert Agent' feature: wrap one-off AI tasks into a reusable 'AI coworker' that can be continuously refined. Agents are AI programs that call tools and autonomously complete multi-step tasks. We see this as a key step for agents moving from demo to practical use — Chinese tech giants are now seriously tackling the experience ceiling of one-shot prompts. The original author demoed a three-step workflow: pull today's bookmarked articles from the ima knowledge base → filter and categorize → generate PPT. Run the task → crystallize it as an expert → refine the settings doc.
Industry view
Supporters see this as a critical step for real-world agent adoption. The 'crystallization' approach addresses the biggest pain point: users rewriting prompts and reconfiguring tools every time, with no way to accumulate experience. WorkBuddy packages task experience into 'scenario capability packs,' letting AI genuinely remember.
Skepticism exists as well:
- Deep ecosystem lock-in: WorkBuddy is tightly bound to Tencent's ima, with limited cross-platform capability.
- It's still essentially preset prompts plus tool calls, lacking true learning ability — refinement still requires humans editing the settings doc.
- Claude Projects and GPTs (OpenAI shipped similar custom-assistant features over a year ago) already exist; Tencent is mostly filling out its product line, not innovating.
Impact on regular people
For individual professionals: common workflows (weekly reports, client follow-ups, information triage) can be packaged as a dedicated AI coworker, eliminating the need to rewrite prompts each time.
For enterprise IT: department-level SOPs can be crystallized, turning veteran employees' experience into AI experts — but it requires dedicated staff to maintain the expert library.
For the consumer market: personal AI assistants move from demo to daily use, but they still depend on specific ecosystems; cross-platform reuse is the next pain point.