DeepSeek Harness hits 1.5M daily users as AI shifts to agents
DeepSeek's open-source agent app Harness hit 1.54 million daily users in three months by doing actual work instead of just talking.
Tools · Source: TradingView (Seeking Alpha)
What happened
DeepSeek says its open-source AI agent application, Harness, reached 1.54 million daily active users by November 4. The growth is rapid. The app launched less than three months ago. At the end of September, it only had roughly 271,000 users. Harness handles complex tasks like coding, file analysis, and research. It does this with very limited human supervision. The company is positioning Harness as a direct open-source alternative to proprietary offerings from OpenAI and Anthropic. DeepSeek disclosed these figures on X. They count people connecting directly to their servers. The numbers reflect company disclosures rather than independently verified adoption.
The computing usage is staggering. DeepSeek reported processing approximately 92 trillion tokens daily. That equals roughly 60 million tokens per active user. Tokens are the basic units of information processed by AI models. Extended agent workflows require repeated model calls. This drastically increases computational demands compared to simple chat queries. The massive token volume raises serious questions about operating costs.
DeepSeek also plans to build a marketplace. Developers will use this to distribute plugins that add functions to the application. Harness already has about 35,000 plugins. DeepSeek says 60 percent of users have created at least one plugin, though their complexity varies. A proposed developer support program is also in the works. This program would include AI usage credits, technical help, and promotion. Further details on the program were not disclosed.
Key facts
- 1.54 million — Daily active users on Harness by November 4
- 271,000 — Daily active users at the end of September
- 92 trillion — Tokens processed daily by DeepSeek
- 60 million — Average daily tokens processed per active user
- 35,000 — Number of plugins already created for Harness
Why it matters
The shift from chatbots to agents is accelerating. Builders are moving past interfaces that just answer questions. They want systems that finish the work. DeepSeek proves there is massive user demand for open-source agents that can grind through hours of coding and research independently. This changes the baseline for what users expect from AI products. Competition is shifting heavily toward applications and developer ecosystems rather than just raw model capabilities.
This transition completely changes the economics of AI infrastructure. Agents consume vastly more compute than traditional chatbots. When users burn 60 million tokens a day on average, operating costs skyrocket. Wider adoption of autonomous agents will dramatically increase global demand for processing capacity. Hardware providers and cloud platforms will see a massive surge in demand. Startups building agents will need to solve the unit economics of infinite model loops.
For builders
Build plugins for the new marketplace
DeepSeek is launching a marketplace for developers to distribute plugins. You can get in early on a platform with over 1.5 million daily users. The proposed developer program will pay out AI usage credits and offer technical help.
Prepare for massive token consumption
Agents run in loops and make repeated model calls to finish tasks. If you build agent applications, your computational demands will grow exponentially. You must optimize your workflows and monitor operating costs before scaling to users.
Leverage open-source agent frameworks
DeepSeek is positioning Harness as an open-source alternative to OpenAI and Anthropic. You can study their ecosystem to understand how to build complex workflows. Use these open tools to build applications without getting locked into proprietary ecosystems.
My take
Chatbots give you answers, but agents actually do the work. DeepSeek is proving that open-source agents can scale incredibly fast. But burning 60 million tokens per user every single day is a massive infrastructure cost. I am watching closely to see if they can afford to keep this running or if the compute bill crushes them.
Original reporting: TradingView (Seeking Alpha). This is my rewrite and opinion.