Software engineers at General Motors' autonomous driving division now spend only about 15% of their time coding, with AI agents handling much of the other 85%, including data analysis, problem triage, and testing. According to Rashed Haq, GM’s VP of autonomous vehicles, this shift has led to roughly three times as many merged pull requests, faster releases, and fewer defects. GM’s strategy focused on redesigning the entire engineering workflow around AI agents rather than just adding coding assistants. The company connected these agents to internal tools and massive data sets using custom Model Context Protocol servers and gave them instruction sets called "skills" to perform specific tasks. This approach allows agents to access vehicle telemetry data, run experiments in parallel, and automate bottlenecks across simulation, road testing, and post-deployment monitoring. With engineers overseeing critical points and agent permissions matching user access, GM successfully increased development velocity while maintaining quality. This comprehensive integration of AI into workflows substantially improved productivity and reduced errors, far exceeding initial expectations.
Iceye, the Finnish space technology firm, has raised €300 million from the Scaleup Europe Fund. CEO and co-founder Rafał Modrzewski shared with Bloomberg that the company envisions going public in Europe in the near future. For further details, visit Silicon Republic.
Runway spent weeks tackling a persistent bug where AI-generated avatars drifted off-center in real-time video generation. Instead of fixing it through backend patches, they created a front-end feature to automatically re-center user images, ensuring stable video output. Ryan Phillips, head of enterprise product at Runway ML, shared insights at VB Transform 2026 on how their real-time generative video technology evolved, highlighting the importance of cross-team evaluation processes and innovative development techniques like distillation and adversarial post-training to optimize performance. Runway’s approach emphasizes turning model limitations into user-friendly features, rigorous infrastructure debugging, and embracing “failure hell” to drive breakthroughs. This new mindset allows creatives to shift from designing single assets to creating dynamic worlds generated in real-time by AI models.
Snowflake has launched Cortex AI Gateway, a centralized platform aimed at regulating how AI agents—including those developed by competitors like Anthropic's Claude Code and Cursor—interact with enterprise data, tools, and models. The gateway features integrations with various identity security providers such as 1Password, Aembit, Linx Security, SailPoint, and Saviynt, creating a collective trust framework for autonomous agents. This move positions Snowflake not just as a data host but as a governance layer controlling AI agent permissions and activities.
The initiative addresses a critical issue: traditional enterprise security models, designed for human actors, are inadequate for AI agents operating at machine speed and across multiple systems. Cortex AI Gateway enforces continuous trust verification for AI agents through dual attribution—logging both agent identity and the human who authorized a task—ensuring task-scoped access and full auditability.
The platform also tackles runaway AI costs by providing IT and finance teams with visibility into AI consumption, attributing spending to specific agents or teams, and enforcing spending limits. This functionality is bolstered by Snowflake’s acquisition of Natoma, which brought advanced identity, policy, and audit features into the ecosystem.
By uniting competing identity vendors under a shared framework, Snowflake aims to prevent new AI silos from forming and foster interoperability. Analyst forecasts predict a trillion-dollar market opportunity in AI agent governance, placing identity verification and control at the heart of enterprise AI strategy. Cortex AI Gateway currently enters public preview, with its partner integrations in private preview, marking a significant step in managing the future of AI agents in business securely and efficiently.
Is AI capable of managing a successful business and directing human employees? A recent experiment challenges this question by giving an AI system its own storefront, three employees, and a $100,000 budget to test its abilities in real-world conditions.
The Model Context Protocol (MCP), an open standard connecting AI agents with global software, has received its biggest update since its launch by Anthropic twenty months ago. Now under the Linux Foundation's Agentic AI Foundation (AAIF), this update revolutionizes MCP into a fully stateless architecture, improves security by enhancing its authentication processes, and introduces a formal 12-month deprecation policy. This transformation enables enterprises to deploy AI agents at scale without the previous challenges of session persistence, allowing seamless operation behind load balancers using modern cloud infrastructure tools like Kubernetes.
Key changes include removing the dependency on persistent sessions, thereby enabling better fault tolerance and scalability, and extending protocol capabilities with new features such as interactive server-rendered interfaces (MCP Apps) and durable long-running asynchronous tasks (MCP Tasks). These improvements allow AI agents to move beyond simple text responses, supporting richer user interfaces and robust handling of complex operations.
Security is fortified by aligning the protocol with OAuth 2.0 and OpenID Connect standards to prevent mix-up attacks, and with enterprise-managed authorization to ensure corporate identity governance. The update is supported by a diverse consortium of technology giants like Microsoft, Google, Amazon, and OpenAI, emphasizing its neutrality and broad industry backing.
Alongside technical enhancements, the introduction of a clear deprecation timeline assures enterprises of protocol stability. MCP is rapidly adopted, evidenced by 250 million weekly SDK downloads, positioning it as a foundational technology for agentic AI workflows in enterprise environments. The new release marks a significant milestone as MCP evolves from an experimental framework into an enterprise-ready standard capable of supporting the next generation of intelligent agents worldwide.
Insights from a founder's journey transitioning from a hands-on technical builder to an effective technical leader guiding transformative change.
Chinese AI startup Moonshot AI has published the full weights of its largest open AI model to date, Kimi K3, a powerful 2.8 trillion-parameter system with advanced features like a one million-token context window. While the model is openly accessible for download and use, enterprises must carefully review the unique Kimi K3 license, which includes commercial restrictions not found in typical open-source licenses. Specifically, companies generating over $20 million annually and providing “Model as a Service” to third parties must negotiate a commercial agreement with Moonshot AI, while large-scale commercial deployments require prominent attribution of Kimi K3 in their products. The license offers a broad grant of rights for internal and non-commercial uses, making it suitable for many enterprises that want to self-host advanced AI capabilities. Developers have reacted positively to the release, praising the accompanying infrastructure and technical reports, though they note the licensing limits and the operational challenges of running such a large model. This release exemplifies the evolving landscape where “open weights” do not always equate to fully open-source software, ushering in complex legal considerations for commercial use. Enterprises should assess their intended use cases carefully to ensure compliance with commercial terms while leveraging Kimi K3’s state-of-the-art AI performance.
At some point, a customer will inquire whether your business is SOC 2 compliant. Here’s a straightforward guide to what SOC 2 compliance means and why it might be important for your SaaS company.
Imagine streetlights equipped with cameras that can identify your car's number plate and even your face. These smart lamp-posts, like those introduced by Conflow Power Group in Warwickshire, come with solar panels for self-powering and additional energy uses. While they promise crime reduction and locating missing persons, concerns about privacy arise as AI may analyze behaviors and movements in unprecedented ways. Big Brother Watch fears a step toward a surveillance state, especially with officials hinting at a future where constant monitoring becomes a norm. The debate continues over who controls the vast data collected and the implications for personal freedom.
