The AI agent startup scene buzzed with activity this week. Alibaba made waves by open-sourcing Qwen2, a model designed to power low-cost AI agents that can operate in multiple languages. This move helps smaller companies build smart tools without huge budgets. Not to be outdone, Chinese rival DeepSeek upgraded its DeepSeek-VL model, enhancing its ability to process both text and images—a key feature for agents helping with design or customer service tasks.

European startups showed strong progress. Belgium’s Conveo secured $5.3 million to expand its AI research coworker that assists scientists in analyzing complex data and summarizing papers. Norway’s 1X announced plans to test its Neo Gamma humanoid robots in hundreds of homes by year’s end, signaling a big step toward practical home assistants.

Funding records tumbled as OpenAI landed a historic $40 billion investment led by SoftBank. The funds will support their new Stargate data center project and an upcoming open model release. Storage tech company MinIO entered the agent space too, launching MCP servers that let users manage AI storage through natural language chats with tools like ChatGPT.

Controversy emerged around Chinese startup Butterfly Effect, which claims its Manus AI agent outperforms OpenAI’s tools in web research tests. Early users praised its ability to handle tasks like resume screening, but others found it getting stuck in loops during testing. Meanwhile, Alibaba staked its claim in the reasoning arena with QwQ-32B, challenging both OpenAI and local rival DeepSeek.

Smaller players made niche advances. Imandra launched CodeLogician, an AI coding assistant that checks its own work for errors—critical for safety-sensitive software. Voice AI saw progress as Groq partnered with PlayAI to release fast, multilingual speech tools supporting Arabic. However, privacy concerns grew as multiple countries moved to restrict certain Chinese AI models over data security fears.

The week highlighted both exciting breakthroughs and growing pains in agentic AI. While startups raced to deliver smarter assistants, questions lingered about reliability and safe deployment—especially as models grow more autonomous.