November 2025 Summaries

Extended Traces is a new feature in the Inngest TypeScript SDK, now available in public beta, designed to enhance observability by incorporating database transaction information, third-party requests, and more into Inngest Traces. This development aims to help developers overcome the limitations of relying solely on error messages or step-level logs for debugging by offering detailed visibility into nested operations such as database queries and HTTP requests. By automatically capturing OpenTelemetry spans within Inngest functions, developers can track every database query, monitor third-party API calls, and visualize nested spans for a comprehensive performance overview. Enabling Extended Traces requires minimal code, and it provides insights into failures in third-party APIs or database queries, measures the performance of individual steps, and correlates spans with workflow steps for thorough end-to-end visibility. As the feature is in public beta, feedback is actively being sought to refine and expand its capabilities.

Context Engineering is a crucial aspect of enhancing AI performance, and Inngest demonstrates its application by building an AI Research Assistant that answers domain-specific questions effectively using a sophisticated context pipeline. The process involves selecting high-quality data sources such as ArXiv, GitHub, and web searches, and leveraging Inngest workflows to retrieve, parallelize, and transform data into relevant contexts. The AI Research Assistant employs the "Orchestrator-workers" pattern, where the context is divided among specialized models—like GPT-4 and Claude—to compress and augment information, ensuring accurate and unbiased responses. This approach highlights the importance of orchestration, data source quality, and the balancing act of rapid data retrieval against API rate limits, ultimately refining the context to optimize the reasoning capabilities of large language models (LLMs). The open-source nature of the project encourages further exploration and experimentation by the community.

Paul Sanglé-Ferrière, co-founder of cubic, discusses the importance of context engineering in AI code review, emphasizing how cubic helps companies like n8n and Granola improve code shipping speed by 28%. Context engineering involves managing the information an AI system accesses to make reliable decisions, crucial for handling large codebases when AI can only process limited tokens at a time. Sanglé-Ferrière shares insights on allowing AI to determine necessary context, selecting appropriate AI models and tools, and addressing production challenges such as rate limits and parallel execution. He highlights the necessity of fast iteration and observability for debugging and improving AI systems. cubic leverages Inngest for durable execution, flow control, and full observability, ensuring reliable context-pulling architecture in production, thus supporting efficient AI workflows and infrastructure critical for successful AI implementations.

Model Context Protocol (MCP) integration for the Inngest dev server introduces AI-assisted workflows, enhancing the testing and debugging of event-driven, asynchronous functions directly within the development environment. This integration allows AI tools such as Claude Code and Cursor to automatically test functions, monitor execution in real-time, and provide detailed error traces, eliminating the need for traditional manual debugging methods. The MCP server offers eight tools for event management, execution monitoring, and documentation access, all embedded within the CLI for offline use. By connecting AI assistants to the local Inngest dev server, developers can streamline their workflow without external dependencies. This shift marks a new era in serverless function development, with future plans for AI-assisted optimization and error diagnosis.