DeepSeek is recruiting specialists to build a multilingual AI search engine and infrastructure for autonomous agents, according to more than a dozen job postings the company published this month. The Bloomberg report details positions for engineers who can construct an engine supporting text, image, and audio inputs.
The listings describe a multimodal system optimized for mobile scenarios. Think screenshot queries and voice clips rather than typed keywords. DeepSeek wants search that delivers answers, not link lists.
The agent play
Separate postings outline infrastructure for AI tools designed to run persistently with minimal human oversight. The company is staffing for training data curation, evaluation frameworks, and platforms capable of hosting numerous agents simultaneously.
One listing for a full-stack developer asks candidates to demonstrate "persistent curiosity about the technological path and development of artificial general intelligence." The phrase appears across multiple job descriptions, making AGI less a marketing flourish and more a stated hiring criterion.
DeepSeek hasn't responded to requests for comment.
Context matters here
The timing is notable. DeepSeek rattled the AI sector a year ago when its R1 model matched OpenAI's o1 on reasoning benchmarks while reportedly costing under $6 million to train. That release triggered a single-day $600 billion wipeout of Nvidia's market value.
Since then, the company has iterated rapidly. V3.1 arrived in August. V3.2 followed in December. A reference to "model1" recently appeared in DeepSeek's FlashMLA repository on GitHub, fueling speculation about a successor model. Reports from The Information suggest V4 may debut around the Lunar New Year in mid-February, with internal tests showing strong coding performance.
Search and agents represent a strategic expansion. Google owns the default search habit. OpenAI has been investing in both areas through ChatGPT's web browsing and the rumored "Operator" agent product. DeepSeek entering this space signals it wants a piece of daily utility, not just benchmark bragging rights.
Why open-source matters for this
DeepSeek has consistently released models under permissive licenses. That strategy created competitive pressure beyond direct performance comparisons. The company's cost edge, achieved partly through algorithmic efficiency rather than raw compute scale, allows aggressive moves into new product categories.
Whether search launches as a standalone product, an API, or embedded in Chinese services like WeChat remains unclear. Baidu already integrates DeepSeek R1 into its search engine. Distribution questions will matter as much as technical capability.
The job postings offer a roadmap. Search finds information. Agents act on it. Combined, they start looking less like a chatbot and more like the AI assistant everyone's been promising for years. DeepSeek appears to be building both pieces simultaneously.
