September 2025 Summaries

Richard Feldman, a software engineer at Zed Industries, discusses the technical decisions behind building the fast code editor Zed, which was designed from the ground up with speed as its primary goal. The company's approach to achieving this speed involved creating custom frameworks and handwritten shaders for different platforms, allowing the editor to render straight to the graphics card and maximize performance. Zed's broader vision goes beyond being a fast editor, aiming to create an open system that captures the entire real-time collaboration process, including conversations and code reviews, with the goal of evolving version control beyond Git. The company also integrates AI features, such as edit predictions and conversational AI, with a focus on customization and low latency, achieved through partnerships like the one with Baseten. Feldman views AI's role in programming as a tool to generate rough drafts and save time, but emphasizes the importance of maintaining control over the final output. Zed's collaboration features enable seamless real-time editing, allowing multiple users to work together on the same document without coordination overhead, making it an effective tool for large-scale collaborative work.

The text presents a detailed account of Baseten's evolution from its inception to its current focus on generative AI, highlighting the founders' journey and the strategic decisions they made along the way. Initially founded by friends with shared interests in machine learning (ML), the company focused on ML infrastructure, particularly inference, which involves serving and scaling models rather than training them. Over time, Baseten pivoted from classic ML models to supporting more complex deep learning models, such as transformers and diffusion models, as they became more prevalent in end-user applications. This shift required the company to enhance its product offering, emphasizing low latency, high throughput, and scalability. Co-founder and CTO Amir discusses the challenges of leading through rapid growth, the importance of maintaining a balance between innovation and focus, and the excitement of enabling non-specialist engineers to tackle AI problems more effectively. Additionally, Amir shares insights on the importance of choosing the right partners during fundraising, likening it to selecting a skilled surgeon for a critical operation.

Baseten has announced a $150 million Series D funding round led by BOND, with new board member Jay Simons, and participation from Conviction, CapitalG, and other investors, marking a significant step in its journey to advance AI infrastructure. The company, which began focusing on AI over 15 years ago, has developed a powerful infrastructure for model inference, offering fast, flexible, and reliable runtimes that surpass competitors by 40-50% in performance. Baseten's platform supports both open and closed AI models, providing developers with control over costs and operational transparency while ensuring high reliability. As AI becomes integral to various industries, Baseten positions itself as a crucial infrastructure partner for companies like Abridge and Sourcegraph, enabling them to deliver superior AI-driven applications. The newfound capital will enable Baseten to further its mission of embedding AI into everyday life and expand its team to support future developments.

Baseten's Multi-Cloud Capacity Management (MCM) system is designed to simplify and enhance AI inference across multiple cloud platforms, providing a universal orchestration layer that treats distributed GPUs as a single, elastic resource. This system ensures 99.99% uptime through active-active reliability, intelligent compute allocation, and routing to achieve the lowest possible latency while complying with standards like SOC 2 Type II, HIPAA, and GDPR. By collaborating with an extensive cloud partner ecosystem, Baseten eliminates vendor lock-in and offers flexible cloud usage options alongside rapid access to the latest GPU technology, such as NVIDIA Blackwell. This infrastructure supports AI engineers by delivering high-performance, production-grade applications with minimal latency and high reliability, while also reducing deployment complexities and costs. Baseten's approach allows customers to operate on a globally reliable infrastructure without the usual scaling challenges, offering a seamless developer experience and future-proof scaling for innovative AI applications.