01 OpenAI hardens Agents SDK with sandboxed, model‑native execution for enterprise agents
OpenAI released an update to its Agents SDK that adds native sandbox execution and what it calls a model-native harness, aiming to help developers build secure, long-running agents that access files and external tools. The change packages execution within a controlled runtime and integrates closer to the model layer, which OpenAI says reduces surface area for unsafe behavior while enabling persistent agent workflows.
The update is pitched at enterprises and teams that need agents to perform multi-step tasks across data and systems; the SDK’s new features are intended to make it easier to manage state, run background processes, and restrict what an agent can access during execution. OpenAI’s blog frames the work as the next evolution of its agent tooling to scale agentic workflows without producing runaway behavior.
For developers this likely means fewer ad hoc wrappers and more built-in controls for long-running agents: sandboxing curbs direct system access, and a model-native harness standardizes how models interact with tools and files. The move echoes a broader industry push to balance agent capability with safety and operational controls.
Takeaways: OpenAI’s Agents SDK now includes native sandbox execution to limit runtime access; a model-native harness standardizes interactions with files and external tools; the changes target enterprise needs for secure, long-running agents.
- Native sandbox execution limits what agents can access at runtime.
- Model-native harness standardizes how agents invoke tools and handle files.
- Targeted at enterprises building persistent, multi-step agent workflows.
02 Allbirds sells brand assets and signals a pivot to AI services as shares jump
Allbirds announced plans to sell its name and assets for $39 million to American Exchange after years of weak profitability and declining sales, according to reporting that summarized the company’s strategic retreat from apparel. The disclosure accompanied a dramatic market reaction: the company’s stock surged in response to the sale and the stated intention to pursue a new direction tied to AI.
Reporting traces Allbirds’ troubles back to slowing sales after a high-profile IPO and notes that the asset sale and pivot amount to a hard reset for the brand. Coverage described the move as emblematic of a speculative cycle in which consumer companies reposition toward AI or infrastructure bets to win investor favor.
Analysts and observers framed the pivot as risky and reminiscent of past market fads, noting that selling core brand assets for a modest cash infusion does not by itself create the infrastructure or customers needed to run AI services. Still, the immediate stock reaction shows how the market rewards narratives tied to AI transformation, even when the operational details remain thin.
Takeaways: Allbirds agreed to sell its name and assets for $39 million to American Exchange; the announcement triggered a large immediate stock increase; commentators compared the move to prior speculative corporate pivots toward trendy technologies.
- Sale of Allbirds’ name and assets valued at $39 million.
- Market reacted strongly, with the stock jumping sharply after the announcement.
- Observers warned the pivot echoes past speculative corporate repositionings.
03 Market watchers call Allbirds’ fashion‑to‑AI pivot a risky, headline‑driven move
Coverage beyond the transaction emphasized the optics: commentators called the shift from apparel to AI compute or services a desperate, stock-boosting maneuver that recalls prior market frenzies. Reporting warned that rebranding as an AI services company does not equate to having the compute, customers, or product roadmap required to compete in that space.
Writers also highlighted the company’s financial backdrop — weak post‑IPO performance and sharply lower sales over recent years — to explain why leadership might pursue a drastic strategic change. That context frames the sale as a tactical move to generate liquidity and a new headline, rather than evidence of an immediately viable AI business.
The tone of this coverage is skeptical but concrete: the pivot raises questions about execution, capital needs, and whether the new strategy can produce durable revenue, not just a temporary market lift.
Takeaways: Critics described the pivot as a speculative, stock-focused maneuver; journalists noted Allbirds’ prior financial struggles as context for the decision; key skepticism centers on whether the company can translate a name-and-assets sale into a real AI-services business.
- Coverage framed the move as a speculative, stock-focused reset.
- Journalists pointed to Allbirds’ weak post-IPO performance and declining sales as context.
- Key doubts focus on execution and the gap between branding and building AI infrastructure.
What moved around the edges
Google brings Gemini assistant to the Mac desktop
Google introduced a Gemini Mac app that brings a floating assistant chat bubble to the desktop and supports an Option+Space shortcut for quick access, letting users share windows and content without switching apps.
The Verge AIGemini 3.1 introduces Flash TTS to broaden expressive speech options
Google published details on Gemini 3.1 Flash TTS, positioning it as the company’s next-generation expressive text-to-speech model for more natural, varied AI voice output.
Google AI BlogBoston Dynamics robots use Google models to read gauges and thermometers
Robots from Boston Dynamics are now using Google’s AI to read industrial gauges and thermometers during inspections, demonstrating practical computer-vision and multimodal model deployments in robotics.
Ars Technica AIHightouch hits $100M ARR after launching an AI marketing‑agent platform
Hightouch reported it reached $100 million ARR, saying the company grew revenue by roughly $70 million in 20 months after introducing an AI agent platform tailored to marketing use cases.
TechCrunch AIMid‑2026 snapshot: open models and the widening open‑versus‑closed gap
An industry commentary laid out expectations for open models through mid‑2026, focusing on how the open-versus-closed model landscape may evolve and what that means for developers and competition.
Interconnects AI