Hugging Face has redesigned its official hf command-line interface to serve both human users and AI coding agents like Claude Code, Codex, and Cursor, automatically detecting when an agent is driving it and adjusting output accordingly. The company reports that on complex multi-step tasks, agents using the hf CLI consume up to six times fewer tokens than those relying on hand-rolled curl or Python SDK approaches. Since beginning to track agent traffic in April 2026, Hugging Face has seen significant usage, with Claude Code alone accounting for roughly 40,000 distinct users and nearly 49 million requests.

The Hugging Face CLI (hf CLI) has been redesigned to better serve both human users and AI coding agents, optimizing output formats for each. AI agents like Claude Code and Codex now account for significant traffic on the Hub, with Claude Code alone handling nearly 49M requests. The updated CLI adjusts its output—simplified and structured for agents, while maintaining rich formatting for humans—to improve efficiency and reduce token usage in multi-step tasks.

Airbnb CEO Brian Chesky plans to launch a new AI lab, though details about its focus remain unclear. Chesky, who has close ties to OpenAI's Sam Altman, will continue leading Airbnb while overseeing the lab’s development. The move puts him in competition with other AI ventures, though he won’t personally lead the lab’s operations.

Airbnb CEO Brian Chesky plans to back a new artificial intelligence lab, joining a wave of Silicon Valley figures dissatisfied with models from existing frontier labs. The venture is expected to focus on areas Chesky has emphasized at Airbnb, such as user interaction and design, though he will not lead it himself and intends to remain Airbnb's CEO. The move puts Chesky in competition with OpenAI, whose CEO Sam Altman he has long advised and helped reinstate after Altman's brief 2023 ouster.

Simon Willison released an alpha version of Datasette Agent with Micropython support, aiming to safely generate and execute Python code. Early tests show promise, with GPT-5.5 unable to bypass the sandbox security measures. The project encourages sponsorship for updates on LLM advancements.

A new alpha release of datasette-agent-micropython enables the Datasette Agent to generate and execute Python code within a sandboxed environment. Early testing suggests the sandbox is holding up well, with GPT-5.5 so far unable to break out of its constraints.

Hugging Face's Reachy Mini robot can now access external tools hosted in public Hugging Face Spaces and called via MCP, letting users add new abilities like weather checks or web searches without editing the app or downloading code locally. Tools run within the Spaces themselves and are activated through profiles, which use a tools.txt file to control which capabilities the robot can use. Users can also publish their own tools for others, expanding beyond the built-in local tools that primarily control the robot's physical movements and expressions.

Reachy Mini can now access tools hosted on Hugging Face Spaces via MCP, allowing users to add new functionalities like weather checks without modifying the app. These tools run remotely in the Space, eliminating the need for local downloads while enabling users to publish and share their own tools. The update expands the robot's capabilities beyond built-in local functions while maintaining security through profile-controlled tool activation.

Meta has built six weatherproof tents, which it calls "rapid deployment structures," outside New Albany, Ohio, to house billions of dollars of AI chips and cut data center construction time in half. The strategy borrows from Tesla's use of tents during Model 3 production and from xAI's reliance on off-grid gas turbines, with Meta's Ohio site powered by 200 megawatts of modular gas turbines. The move comes as Meta plans to spend up to $145 billion on data centers and other capital expenditures, a figure that has weighed on its stock.

Meta has adopted a strategy pioneered by Tesla, constructing rapid deployment data centers in tents to accelerate construction timelines and reduce costs. These tent structures, located outside New Albany, Ohio, house AI chips and are powered by nearby modular gas turbines. The move is part of Meta's broader $145 billion investment in data centers and capital expenditures amid the competitive AI race.

Charity Majors argues that both AI enthusiasts and skeptics on software teams are responding to genuine existential threats: enthusiasts fear being outpaced by competitors who embrace AI, while skeptics warn that shipping code faster than engineers can comprehend erodes reliability and institutional knowledge. She contends there is no natural feedback loop linking the two camps, framing the task of designing such mechanisms as both a leadership and engineering challenge aimed at restoring a shared understanding.

AI enthusiasts push for rapid adoption to stay competitive, while skeptics warn of unsustainable practices that erode reliability and trust. Both perspectives highlight real risks—falling behind technologically versus collapsing under technical debt. The challenge lies in bridging the gap between these views through better feedback loops and organizational design.