May 2026 Summaries

Neon is expanding its capabilities beyond its foundational serverless Postgres database to create a comprehensive backend platform aimed at the agentic era, characterized by AI-driven applications and agents. The platform will include essential backend primitives such as authentication, data API, object storage, compute, and an AI gateway, all built with a focus on providing a stable and scalable infrastructure rather than relying on "magic" solutions. This expansion is in response to the increasing demands of AI-driven development, which necessitates robust infrastructure capable of handling high-speed operations and potential errors from automated agents. By leveraging its partnership with Databricks, Neon has enhanced its engineering team and capabilities, enabling the development of these new features without diverting attention from its core Postgres product. The new primitives are designed to integrate seamlessly with existing Neon projects, offering additional functionality without requiring significant changes from current users.

Neon's lakebase architecture is engineered to maintain resilience against cloud failures, addressing the challenges posed by the increased demands of agent-driven workloads. This architecture leverages a stateless Postgres compute model, separating compute and storage to enhance availability without requiring costly hot standbys or long recovery times. Neon's approach includes a high availability design that utilizes zone-redundant storage and a unique control plane that functions as a data plane for critical operations, ensuring reliability in database startups and management. The platform employs a cell-based architecture to limit the impact of failures, allowing for scalable regional growth while containing potential disruptions. A rigorous testing regime involving failure simulation and injection ensures robust system reliability, with the goal of maintaining high availability standards for all databases. Continuous measurement of service level indicators and objectives helps track and improve the system's performance, aiming for best-in-class reliability and user trust in Neon's database services.

Neon has introduced "neon-for-agent-platforms," a companion skill designed to manage and operate fleets of Postgres databases on agent platforms, complementing the existing "neon-postgres" skill. The skill addresses the unique challenges of agent platforms that require provisioning and managing numerous databases, offering solutions such as dual-org fleets, project transfer, per-tenant provisioning, and a consumption API. It provides guidance on creating and managing databases at scale using real-world patterns from major clients like Replit and Retool. The skill includes sample TypeScript scripts for practical implementation, focusing on control-plane patterns like dual-org models, per-tenant provisioning, and compound checkpointing. It aims to streamline workflows by teaching agents to handle database operations efficiently, reducing the need for repeated problem-solving among large agent platform customers. Users are encouraged to install the skills in their preferred editor, enabling AI assistants to perform platform-level tasks with appropriate API calls and patterns. The open-source nature of the project invites user feedback and collaboration, aiming to refine the skill further based on community input.

Neon has significantly enhanced write performance, achieving up to a 5x improvement for write-heavy workloads by disabling full-page writes (FPW) in Postgres, which are redundant due to Neon's storage architecture. This architectural change separates compute and storage, enabling the offloading of tasks from Postgres compute to distributed storage, which is not possible in traditional Postgres deployments. The move, known as image generation pushdown, allows Neon's pageserver to generate full-page images based on actual changes, reducing network traffic by 94% and improving scalability and read performance. Benchmarks and real-world production tests have shown substantial improvements in transaction throughput and read latencies, with WAL generation dropping significantly. The change has been rolled out seamlessly across all Neon databases globally, marking a significant step in leveraging the flexibility and performance potential of Neon's lakebase architecture.