Back in the Harness: DeepSeek continues its pivot towards agentic AI

Chinese artificial intelligence company DeepSeek is venturing into a new battleground beyond large language models, launching a developer preview of its long-anticipated Harness – a software framework that helps developers turn AI models into autonomous agents. The release on Thursday marks a strategic pivot as DeepSeek moves to build foundational digital scaffolding for AI agents – systems capable of using AI models to operate external software, run code, and complete complex jobs on their own. Following earlier beta testing, taking part developers noted that DeepSeek Harness departed from conventional software tool kits by giving programmers greater control over how their AI agent thinks and acts. Each core element was delivered as a modular plug-in that developers can swap out or combine. Tools like DeepSeek Harness are rapidly becoming an important layer in AI systems, shifting the global competition from model intelligence to one that focuses on how seamlessly an AI agent can connect to real-world software. DeepSeek hired former Jane Street engineer Cui Tianyu in March to join its newly formed harness group, amid the surging popularity of powerful AI agents like US rival Anthropic’s Claude Code. To cater to different enterprise and developer workloads, DeepSeek Harness provides four operational settings. These include a standard mode for general tasks, a code-focused mode that lets the AI write code to command multiple applications simultaneously, a creative mode that allows the system to experiment with custom tools, and a minimal mode for isolated testing. DeepSeek’s highly customisable approach stands in sharp contrast to San Francisco-based Anthropic, which built its Claude Code tool as an integrated, ready-to-use product. Chinese developers, on the other hand, are increasingly using open, adaptable infrastructure to drive global adoption across their own software ecosystems. The new product comes as DeepSeek is preparing to significantly increase the cost of using its AI models through its application programming interface services. Starting Monday, the company will introduce peak and off-peak pricing tiers while raising fees across its flagship V4 models. Under the new pricing structure, cache-hit input costs for its premium V4-Pro model will jump from 0.36 US cents per million tokens to 2.20 US cents during off-peak hours and 4.40 US cents during peak hours – a sixfold and 12-fold increase, respectively. DeepSeek earlier this week released DeepSeek-V4-Pro-0813, an updated version of its flagship model, leaving some developers disappointed by its overall capabilities and pricing – but impressing researchers in niche areas like cybersecurity.