5-min read · Curated daily by an AI Systems Architect
Focus: Agentic Workflows · AI Coding Tools · Embodied Intelligence
1. Cognition AI Raises $1B at $26B Valuation — Devin Doubles in 9 Months
【Technical Core】
Cognition AI, the maker of autonomous AI software engineer Devin, has raised over $1 billion at a $25 billion pre-money valuation ($26B post-money). The round represents a 2.5x jump from September 2025, when the company was valued at roughly $10 billion. While Claude Code and Codex dominate the AI coding headlines, this raise signals that the AI coding agent market is far from a two-horse race.
【Why It Matters】
A $26B valuation for a 3-year-old AI coding startup is unprecedented. It confirms that autonomous software engineering agents are viewed as a platform play, not a feature. Cognition's path — Devin as a standalone AI engineer rather than a terminal plugin or IDE extension — represents a fundamentally different bet on how AI will reshape software development. The capital will likely fuel deeper model capabilities and enterprise deployment.
2. First npm Supply-Chain Attack Targets Claude AI Tool Environment
【Technical Core】
A malicious npm package named "mouse5212-super-formatter" has been caught exfiltrating files from Claude Code's dedicated upload directory to a threat actor-controlled GitHub repository. This is the first documented supply-chain attack built specifically to target an AI coding tool's environment, not generic developer machines. The malware specifically scanned for and stole files from Claude's user data directory.
【Why It Matters】
As AI coding tools become integral to developer workflows, they become high-value attack surfaces. Claude Code stores conversation history, project context, environment variables, and sometimes API keys in its local directory. This attack demonstrates that threat actors are now specifically targeting AI toolchains — a new frontier in software supply-chain security. Developers using AI coding agents need to apply the same dependency hygiene they'd use for any production pipeline.
🔗 AI Weekly · BitNewsBot
3. Robinhood Lets AI Agents Trade Stocks Autonomously via MCP
【Technical Core】
Robinhood has launched agentic trading capabilities — users can now connect AI agents like Anthropic's Claude or OpenAI's ChatGPT to dedicated investment accounts via the MCP (Model Context Protocol) standard. The agents can autonomously execute stock trades on a separate sub-account with guardrails. Robinhood also introduced an AI agent credit card with 3% cash back, enabling agents to make purchases on users' behalf. The launch has US broker regulatory approval (FINRA).
【Why It Matters】
This is a landmark moment for agentic commerce. AI agents have been moving from information retrieval to action execution — coding, email drafting, calendar management — but financial transactions represent a new level of trust delegation. The use of MCP as the integration standard is also notable: it reinforces MCP as the de facto protocol for connecting AI agents to real-world services. The "AI agent credit card" concept suggests a future where agents manage not just investments but household spending.
4. Unitree G1 Hits 100,000 Pre-Orders at $16,000 — Humanoids Go Mass Market
【Technical Core】
Chinese robotics company Unitree announced 100,000 pre-orders for the G1 humanoid robot, priced at $16,000 — cheaper than a Honda Civic. The company also released a video of the G1 successfully performing a side flip, demonstrating rapidly advancing dynamic motion control. Unitree ranked first globally in humanoid robot shipments last year.
【Why It Matters】
100K pre-orders at $16K signals that humanoid robots are crossing from lab curiosity to consumer product. This is the Tesla Model 3 moment for humanoids — the price point where adoption curves bend upward. The side flip demo is also meaningful: dynamic whole-body motion control at this level was science fiction just two years ago. Combined with voice command capabilities demonstrated earlier this month, the G1 is becoming a general-purpose platform.
🔗 Robot News Today · Beijing Post
5. Apptronik Raises $350M from Google Ventures — Mercedes Expands Apollo Warehouse Ops
【Technical Core】
Austin-based Apptronik closed a $350 million Series B round led by Google Ventures. Concurrently, Mercedes-Benz announced it is expanding Apollo humanoid robot deployment from pilot programs to full-scale warehouse and manufacturing operations. The Apollo robots are performing material handling, kitting, and assembly-line support tasks in Mercedes facilities.
【Why It Matters】
This is the industrial counterpart to Unitree's consumer push. Apptronik + Mercedes represents one of the most concrete examples of humanoid robots in real manufacturing environments — not demos, but production-line work. Google Ventures' lead investment also signals Big Tech's conviction that the humanoid robotics supply chain is maturing beyond R&D. The combination of consumer (Unitree) and industrial (Apptronik) milestones on the same day paints a picture of embodied AI crossing the chasm simultaneously on both fronts.
🔗 Robot News Today · Figure News
6. Shanghai Launches "Ge Wu" Embodied AI Platform — One Codebase, 100+ Robot Types
【Technical Core】
Shanghai's National and Local Co-Built Humanoid Robotics Innovation Center unveiled "Ge Wu," an embodied AI simulation platform that uses a universal reinforcement learning framework to train over 100 different robot types from a single codebase — no additional programming required per robot model. Shanghai is also partnering to establish a humanoid robot subcommittee under ISO/TC299 (Robotics Technical Committee) to pursue international standards. The city has built China's first heterogeneous humanoid robot training facility, currently training 100+ robots with a target of 1,000 by 2027.
【Why It Matters】
The "single codebase, 100+ robots" claim is the key breakthrough. If validated, it solves one of embodied AI's hardest problems: the sim-to-real transfer gap that forces separate training pipelines for every robot morphology. The ISO/TC299 standards push is equally strategic — China is moving early to shape the international regulatory framework for humanoid robots, much as it did with 5G. The training facility target (1,000 robots by 2027) suggests the infrastructure is being built for industrial-scale embodied AI deployment.
Top comments (0)