July 2025 Summaries

The text explores the evolving landscape of software engineering, emphasizing the distinction between "disposable" and "durable" code. Disposable code, often used for quick, experimental purposes like design mocks and prototypes, has gained traction with advancements in AI tools, enabling rapid generation and iteration. In contrast, durable code underpins critical systems like banking and healthcare, requiring reliability, predictability, and extensive maintenance. The cost of software largely hinges on its maintenance, with durable code demanding investment in testing, observability, and risk management to ensure trust and stability. While AI aids in both domains, it's particularly transformative in reducing the time and complexity of tasks traditionally associated with durable code, such as instrumentation and incident analysis. The future of software development will likely continue to balance these two modalities, with disposable code serving as a skill set and durable code remaining a profession, as the latter's reliability and trustworthiness are indispensable for high-stakes applications.

Harnessing the power of custom GPTs, the author describes how they have integrated generative AI into their workflow as an engineering leader to enhance communication, reflection, and strategic planning. By using AI tools to draft weekly updates, conduct monthly reviews, and compile quarterly business reports, they have shifted from labor-intensive processes to more efficient and insightful practices. These tools facilitate structured conversations that help maintain a focus on impact over mere activity, enabling the author to communicate effectively with leadership and align team efforts with broader business goals. Despite these benefits, the author emphasizes that AI cannot replace the nuanced aspects of human management, such as performance evaluation and understanding individual contributions. The suggestion is made for engineering leaders to create their own custom GPTs tailored to their specific needs and communication style, promoting deeper reflection and more intentional interactions.

AWS Summits are annual conferences where new products and technologies are announced, with the 2025 theme focusing heavily on AI. The New York Summit highlighted AWS' commitment to AI with their Amazon Bedrock AgentCore, a foundational set of APIs and services designed to enable Agentic AI securely and scalably. This includes a serverless runtime for executing tasks, isolated data environments, and a CodeInterpreter for writing and running code in popular languages. AWS also introduced support for vector storage querying in S3 and showcased their Kira IDE and Amazon Q for enhanced development practices. Thomson Reuters presented their use of AWS Transform for platform migrations, emphasizing empowerment rather than developer replacement. Honeycomb MCP, now available on the AWS Marketplace, offers advanced data analysis capabilities, while Anthropic's talk on Claude Code highlighted the evolution of AI agents from assistants to collaborators. The summit illustrated a clear shift towards AI as a fundamental part of software development, with AWS positioning its offerings to support this transition.

AI-assisted coding can significantly boost productivity, but its effectiveness largely depends on the qualities of the codebase, such as organization, consistency, and clear documentation. In legacy projects, which constitute most valuable software development, AI can be slower if these conditions are not met. To harness AI's potential, codebases should have well-documented best practices, consistent style, strong type checks, and automated testing frameworks, allowing AI to make meaningful contributions. This involves maintaining a style guide, automating style checks, and implementing robust feedback loops for code verification. Although AI can introduce and enforce consistency more efficiently than manual efforts, it requires explicit instructions and sufficient context to execute tasks correctly. Developers are encouraged to document code architecture and establish guidelines that AI can follow, which leads to a more manageable and humane codebase, benefitting both AI and human developers in the long term.

Honeycomb has been recognized as a Visionary in the 2025 Gartner Magic Quadrant for Observability Platforms, highlighting its commitment to helping engineering teams manage increasingly complex systems with AI integration. As the software development landscape becomes more intricate, Honeycomb addresses challenges such as fragmented tech stacks and rising telemetry volumes by offering an event-based pricing model that encourages high-cardinality telemetry without unpredictable costs. The platform's capabilities, including fast querying, anomaly detection, and AI-enhanced developer tools, empower teams to navigate and troubleshoot complex environments effectively. Honeycomb's strategic focus on AI, demonstrated by the acquisition of Grit and innovations like the Query Assistant, aims to simplify OpenTelemetry adoption and improve developers' productivity. It is designed for forward-thinking teams including platform engineers and DevOps, positioning itself as a crucial tool for future-proofing observability strategies in an AI-driven world.

Honeycomb has launched the public beta of its Hosted Model Context Protocol (MCP), providing seamless integration with AI tools like Cursor, Visual Studio Code, and Claude Desktop, and available on the AWS AI Agents marketplace at no cost for all Honeycomb plans. This solution aims to address challenges associated with traditional MCP servers, such as cumbersome API key management and deployment issues, by offering user-friendly features like one-click integration and curated tools for managing context width. Notably, the Hosted MCP includes functionalities such as trace view, column search, and a multi-signal query tool, enabling AI agents to efficiently navigate data and perform tasks like critical path detection and SLO remediation. The platform's capabilities have been showcased in real-world scenarios, such as identifying latency spikes and optimizing infrastructure cost reporting, highlighting its potential for enhancing software development and operations through AI-assisted development. Honeycomb encourages users to explore these features by enrolling in Honeycomb Intelligence and participating in upcoming webinars on AI-driven development.

Honeycomb has announced the availability of its Hosted Model Context Protocol (MCP) Server in the AI Agents and Tools category of the AWS Marketplace, allowing customers to discover, purchase, and deploy AI agent solutions with their AWS accounts. This MCP Server aims to integrate observability data into AI agents, enhancing the software development lifecycle by enabling real-time feedback loops in integrated development environments (IDEs) and reducing developer cycle times. Christine Yen, CEO of Honeycomb, highlighted that the platform's capabilities help identify inefficient AWS usage and analyze cloud utilization, demonstrating its practical value. Features of the Honeycomb MCP include full query support for telemetry data, AI context window optimization, and enterprise deployment compatibility through SSO OAuth. The centralized purchasing system via AWS accounts offers customers control over licensing and payments, expediting procurement processes and fostering AI innovation.

Investing in adding rich context to telemetry significantly enhances debugging and collaboration during incidents, with numerous improvements achievable without altering application code or requiring cross-service coordination. By leveraging OpenTelemetry resource attributes and CI/CD systems, teams can embed valuable context into their telemetry data, such as service tier, team ownership, and deployment details, which facilitates efficient incident management. The implementation of a consistent naming convention for attributes across services and the development of centralized libraries for telemetry abstraction can prevent telemetry drift and ensure standardized practices within organizations. Incident reviews serve as pivotal learning opportunities for identifying gaps in observability and refining telemetry practices, offering insights into system vulnerabilities and the tools required for effective debugging. Ultimately, a proactive approach to integrating contextual information into telemetry can provide significant benefits with minimal effort, prompting users to explore additional enhancements that improve their operational efficiency.

Distributed systems, while offering scalability and flexibility, introduce complexity in troubleshooting due to the intricate network of microservices, databases, and APIs. OpenTracing emerged as an open-source, vendor-neutral API to standardize distributed tracing, allowing developers to instrument their code once for interoperability across different tracing tools. Launched in 2016 under the Cloud Native Computing Foundation, it set the stage for modern tracing standards. However, the need for a unified observability framework led to the creation of OpenTelemetry (OTel) in 2019, which integrates the functionalities of OpenTracing and OpenCensus, providing a comprehensive solution for metrics, logs, and traces collection. OpenTelemetry's compatibility with existing OpenTracing systems allows for gradual migration, aiding organizations in enhancing their observability standards without immediate re-instrumentation. Honeycomb utilizes OpenTelemetry to offer advanced insights into distributed systems, helping identify bottlenecks and anomalies. As OpenTracing is now archived, teams are encouraged to transition to OpenTelemetry for future-proof observability.

Understanding the flow of requests through distributed, cloud-native systems is increasingly challenging for modern software engineers. Tracing is a crucial tool in this context, offering a detailed timeline of a request's journey through various components like microservices, which helps in debugging, performance optimization, and understanding user impact. OpenTelemetry is a widely used framework for collecting and exporting trace data, simplifying instrumentation across programming languages. Traces consist of spans, each representing a single operation within a system, and context propagation ensures these spans are connected, providing a comprehensive view of a request. Traces, when combined with logs and metrics, offer powerful insights into system performance and reliability. Advanced techniques such as trace analytics and sampling further enhance the ability to analyze high volumes of trace data efficiently, while service maps and visualizations provide an overview of system dependencies and latencies. Honeycomb, an observability platform, enhances tracing with features like fast querying and integration with OpenTelemetry, allowing teams to maintain precise and context-rich tracing practices.

Honeycomb's approach to handling sampling traces and logs in observability is a standout feature that often surprises new users with its capability to maintain data accuracy while optimizing cost efficiency. Since 2016, Honeycomb has been automatically correcting all data for sample rates, ensuring that each datapoint in charts is adjusted without requiring user intervention, while also allowing detailed examination through Usage Mode. This includes the innovative use of tail-based sampling, facilitated by the Refinery proxy, to prioritize keeping error and slow trace data, which is crucial for customer support and service scenarios. Honeycomb's EMA Dynamic Sampling further enhances data prioritization by adjusting sample rates dynamically to achieve desired target rates or throughput levels. Additionally, the Enhance feature provides Data Lake-like functionality to retrieve important dropped data, ensuring comprehensive data analysis. The platform also offers educational resources through its Honeycomb Academy course for users to gain deeper insights into these advanced sampling techniques.

At the LDX3 event in London, discussions among engineering leaders highlighted the growing importance of observability in balancing the pressures of rapid delivery and system resilience, especially as AI code generation raises expectations for speed. The sessions underscored that the traditional tradeoff between speed and stability is outdated, with successful teams now optimizing for both by integrating observability as a cultural practice rather than merely a toolset. Challenges such as aligning cross-functional teams on resilience, linking technical metrics to business outcomes, implementing smart guardrails for AI-generated code, and reframing technical debt as a business risk were identified as critical areas of focus. Observability is increasingly recognized as a strategic leadership narrative, requiring investment not just for incident resolution but for informed decision-making and learning acceleration. The event emphasized that thriving teams will be those that effectively balance speed and stability while maintaining a strong focus on observability.

OpenTelemetry's integration with Claude Code allows users to capture detailed metrics and logs, such as token usage and session length, which can be monitored using Honeycomb. By setting specific environment variables and creating a separate environment in Honeycomb, users can track Claude Code's telemetry data effectively. The process involves configuring Claude Code to export telemetry data to Honeycomb, and subsequently using this data to create monitoring dashboards with board templates. These templates help in visualizing various metrics, including tool success rates and session activities, and can be customized with minimal manual adjustments. The setup encourages efficient monitoring of Claude Code's performance, enabling users to identify outliers and analyze API response times using Honeycomb's features like BubbleUp Anomaly Detection.