Which software is AI "killing"?

AI is swallowing standard tasks — and fast

AI is already replacing discrete software features that do rote, repeatable work. Overseas, OpenAI has pushed ChatGPT into document and spreadsheet execution; Anthropic has been embedding agents into legal and financial workflows. In China, startups such as MiniMax and Kimi have folded Office-style generation and editing — Excel, PPT, document analysis — into conversational products. The result? Office users are switching from clicking menus to issuing prompts. A young operator interviewed said she "rarely opens WPS (金山办公)" anymore — AI drafts a full project PPT from a simple outline in minutes. NVIDIA CEO Huang Renxun (黄仁勋) has even argued that traditional apps may give way to AI agents as the dominant software form. So which apps survive when the mundane becomes a baseline skill of large models?

Developers and single-purpose tools feel the heat

AI is compressing software development cycles and lowering barriers to entry. It has been reported that Alibaba Cloud (阿里云) uses AI-assisted code generation for roughly 40% of internal workloads, and that Tencent (腾讯) reached a similar level by mid‑2025; Baidu (百度) is reportedly around the same mark. Practitioners say features that once required weeks of engineering can now be delivered in days with AI-assisted tooling, shrinking demand for mid-sized custom-development shops and prompting layoffs. Simple, narrowly focused tools — older code editors like Sublime Text and HBuilderX, lightweight survey and form services, basic diagramming apps — are singled out as highest risk because their functions map cleanly to promptable steps. By contrast, entrenched vertical platforms with proprietary, hard-to-access data — Bloomberg, LexisNexis, Epic, Procore — look safer for now. And yet disruption and partnership can arrive in the same breath: it has been reported that Anthropic’s push into legal agents first knocked down multiple software stocks, then rebounded after the company announced enterprise plugin partnerships with incumbents.

Strategy, geopolitics and the near-term future

The coming shake‑out won’t be purely technical; it will be strategic and geopolitical. Export controls and supply‑chain tensions around advanced chips affect how quickly different players can scale compute-heavy models, so hardware access will shape the pace of change. But even with constrained GPUs, agent infrastructure, APIs and ecosystem partnerships are advancing — meaning software companies face two clear paths: embed into AI platforms as indispensable domain components, or risk being subsumed as replaceable features. Which will win? For many vendors it will come down to data moats, integration depth and speed of execution. The large‑model era has begun a software reordering; now the question is whether incumbents adapt quickly enough to survive it.