An open-source tutorial uses 3 Agents (AI programs capable of autonomously calling tools and completing tasks step-by-step) + Tongyi Qianwen to successfully replicate OpenAI's $200/month Deep Research—the barriers to Agent applications are far lower than we thought.
What this is
OpenAI launched Deep Research earlier this year, capable of automatically planning, searching, and writing long research reports. It operates on two layers: the underlying DeepSearch performs a "search → read → reason" loop to find answers, while the upper layer adds outline planning and chapter-by-chapter generation. This tutorial built a three-stage pipeline using LangChain + Tongyi Qianwen qwen-plus + DuckDuckGo: planner_agent breaks down keywords, search_agent conducts concurrent searches and summarizes, and writer_agent consolidates everything into a Markdown report. The architecture is not complex, and the code runs.
Industry view
We are seeing Big Tech's Agent features rapidly reverse-engineered by the open-source community, and this is no isolated case. The core of Deep Research—planning + searching + integrating—can be replicated by anyone familiar with LangChain within a week. However, we must note that the chasm between a running demo and a usable product is significant: qwen-plus's reasoning ability has a clear gap compared to o3, DuckDuckGo's search quality falls far short of professional databases, and the reports' accuracy and depth still lag behind the paid version. As some developers bluntly state, such replicas are better suited for learning Agent architectures; truly replacing manual research is still a long way off.
Impact on regular people
For enterprise IT: Building internal research assistants using domestic models + open-source frameworks keeps costs controllable, but we advise against expecting them to replace analysts yet. For individual workers: Those who can write Agents are automating repetitive tasks like "searching for information and writing reports," which delivers a tangible efficiency dividend. For the consumer market: Deep Research-style products will quickly become standard across major LLMs, and paywalls won't hold for long.