May 2026 Summaries

At PostHog, product marketing integration was initially challenging due to a lack of experience collaborating between product teams and marketers, but the necessity of such collaboration for successful product launches became quickly apparent. The author, a product marketer, took proactive steps to establish their role by initiating contact with product managers, participating actively in team Slack channels, and joining both remote and in-person meetings to align marketing efforts with product development. They emphasize the importance of maintaining open communication with product managers through shared planning documents and regular growth reviews to identify marketing opportunities and gaps in product performance. Utilizing AI tools like Claude for automating routine tasks and gathering insights from Slack channels enhances efficiency, allowing the marketer to focus on strategic collaboration and effective communication of product updates to users.

PostHog is focusing on integrating more AI-powered features into its platform, aiming to create proactive, self-driving products like the beta version of PostHog Code, which is designed to surface solutions and improve over time. The company plans to train models on PostHog's data to enhance existing products and develop new ones, such as advanced session replay analysis and synthetic user testing, to anticipate user behavior and improve product outcomes. While PostHog intends to use anonymized data from its US cloud instance for model training, with the option for users to opt out, it emphasizes transparency and communication about these changes, highlighting that the aim is to enhance product quality rather than monetize user data. The opt-out approach is necessary to gather sufficient data for effective model training, and users can manage their participation through PostHog's settings.

Customer-facing analytics are statistics displayed to users, enhancing their interaction with a product, such as usage dashboards, live visitor counters, and public leaderboards. These analytics rely on a fast, authenticated API infrastructure to deliver real-time data seamlessly into user-facing interfaces, exemplified by companies like Vercel, Booking.com, Product Hunt, HogHero, and GitHub. Vercel offers instant data updates on its analytics dashboard, while Booking.com integrates live metrics directly into its landing pages to boost conversions. Product Hunt uses a leaderboard format to engage its community, HogHero enriches customer support interactions by embedding relevant data into existing tools, and GitHub's contribution graph fosters community participation. These platforms demonstrate the importance of integrating timely and relevant analytics where decisions are made, ensuring data is trustworthy and embedded within the product's user experience. PostHog's Endpoints feature facilitates building such analytics by transforming trusted data insights into fast, versioned APIs that can be easily integrated into any customer-facing platform without needing extensive backend infrastructure.

As the integration of AI into customer workflows continues to grow, the role of support is evolving to become more human-centric and technically skilled, rather than being replaced by AI-driven solutions. The text argues against using AI as a barrier to human interaction in customer support, emphasizing that while AI can efficiently handle simple tasks and provide context, it lacks the capability to address more complex issues that require human understanding and empathy. The future of support lies in a symbiotic relationship where AI aids in gathering and presenting information, allowing support engineers to focus on meaningful interactions that address customer needs and foster loyalty. This approach not only enhances customer satisfaction but also positions support as a valuable revenue-generating component of a business rather than merely a cost to be minimized.

After migrating the local evaluation endpoint for feature flags from Django to Rust, significant improvements in performance and resource efficiency were observed, with latency reduced from 40ms to 4ms, CPU usage decreased by 24 times, and memory usage reduced by 56 times. The transition involved porting the endpoint logic, ensuring compatibility with Django's behavior, particularly in authentication and ETag handling, and splitting the Rust service into two fleets to manage distinct workloads. The rollout was executed incrementally, starting with internal SDKs and gradually shifting external traffic to ensure stability, revealing some bugs that were subsequently addressed. As a result, the Rust service now efficiently handles the same workload with significantly fewer resources, demonstrating the benefits of moving the feature flags local evaluation endpoint to Rust.

Vibe coding, a term coined by Andrej Karpathy in 2025, refers to building software using AI to write most of the code based on plain language descriptions, allowing for rapid development often by individuals or small teams. Despite the speed and cost-effectiveness of vibe coding, integrating analytics from day one is crucial to make informed decisions, especially as AI-generated apps can ship features quickly but may not automatically align with user needs. This guide emphasizes the importance of three types of analytics: product analytics to track user interactions within the app, web analytics to understand traffic sources and user behavior before they enter the app, and LLM observability to monitor AI feature performance and associated costs. PostHog is highlighted as a comprehensive tool that covers all these analytics aspects, providing features like autocapture, integration with AI workflows, and cost-effective plans, which are particularly beneficial for early-stage startups using vibe coding. The guide underscores the necessity of analytics to bridge the gap between having a minimum viable product and creating a marketable product, especially in the fast-paced environment of AI-driven development.

In 2026, PostHog's Ops & Finance team embraced an "AI-native" approach to hiring, prioritizing candidates who inherently integrate AI tools into their workflows over those with traditional experience. This shift reflects a broader trend in which AI proficiency is seen as a fundamental job skill, transforming roles in operations, talent management, and finance by significantly enhancing productivity and efficiency. For example, AI-native ops managers automate task management and generate employment contracts, talent partners use AI to streamline candidate feedback and analyze hiring trends, and finance managers employ AI to automate financial processes and improve cash flow management. Despite the advantages, there are risks, such as over-reliance on AI tools and the need for experienced oversight in areas like accounting. PostHog's hiring process is designed to identify genuinely AI-native candidates by evaluating their ability to build and implement AI-driven workflows, with a focus on practical application and adaptability across various tools. This approach has allowed PostHog to operate with a leaner team, as AI-native hires offer increased efficiency and resourcefulness, potentially reshaping the future of work at the company.

PostHog has experienced significant growth since 2021, yet its product marketing team remains small, focusing on efficiency and adaptability rather than rigid frameworks. This approach, humorously termed "minimum viable product marketing," emphasizes rapid and effective launches, allowing the team to ship more frequently than typical companies. The process involves crafting clear and concise announcements, leveraging storytelling centered around conflict, and choosing the right channels for communication, such as Hacker News and newsletters, rather than traditional PR methods. Internal communication is prioritized to ensure all team members are informed, while sales content is developed post-launch to reflect genuine user experiences. While this method may seem chaotic, it has proven effective so far, eschewing complex marketing strategies in favor of straightforward, authentic storytelling.

MCP servers have become a crucial tool for startups, enabling seamless integration and automation across various workflows by allowing agents to perform tasks such as debugging, shipping code, understanding user behavior, planning work, and answering data-related questions directly within an editor. The core advantage of MCP servers lies in their composability, enabling actions like pulling error traces, checking deploys, querying logs, and editing code with minimal context-switching. Popular servers include PostHog for product analytics and error tracking, GitHub for code management, and Vercel for deployment monitoring. The setup involves selecting appropriate servers based on the team's stack and workflow needs, with a recommendation to start with a few key servers, ensuring they are either official or maintained by a reputable team to ensure security and reliability. The use of OAuth for authentication and read-only modes is emphasized to safeguard against potential security issues. PostHog, among other MCPs, is highlighted for providing a comprehensive solution due to its ability to integrate multiple tools under one connector, facilitating efficient and informed decision-making across various startup operations.

Over the past few years, the evolution from traditional Search Engine Optimization (SEO) to Artificial Intelligence Optimization (AEO) has been significant, with the latter involving optimizing content for AI models to increase visibility in AI-generated recommendations. The shift is evidenced by PostHog's remarkable growth in LLM-referred traffic, highlighting the effectiveness of AEO strategies. Despite this success, challenges remain, such as ensuring all product offerings are equally recognized by AI models. The author emphasizes maintaining foundational SEO practices while adapting content for AI consumption, advocating for a focus on quality and specificity in content creation. They stress the importance of being honest in evaluating AEO results, integrating user feedback directly into strategy, and fostering peer learning to navigate the uncertainties of this emerging field. The narrative underscores the need for adaptability and strategic content management in the face of evolving AI technologies.

Last month, AI agents deployed to PostHog projects effectively addressed 6,124 errors by resolving 4,063 issues, suppressing 1,751, and routing 310 to the appropriate teams, showcasing the potential for self-driving products to manage routine tasks. The agents performed triage actions that fell into three categories: resolving issues, suppressing known noise, and rerouting errors to the correct owners. This process highlighted the need for distinct tools for different tasks and demonstrated that errors serve as starting points for deeper investigations into surrounding events and user data. The effectiveness of AI-driven triage was particularly notable in handling bulk operations like suppressing third-party noise or routing platform-specific crashes, which previously required extensive manual effort. The insights gained from this deployment suggest a need for better business-surface filtering and the importance of focusing on distinct workflows for resolution, suppression, and routing to enhance efficiency and maintainability in error management.

PostHog Code, now in beta, is a desktop app designed to enhance product development by running coding agents on top of product data, aiming to achieve a self-driving product development experience. It autonomously handles minor tasks like bug fixes, UX improvements, and conversion tweaks through tools, skills, signals, memory, and evaluation, allowing engineers to focus on more strategic work. By utilizing PostHog's infrastructure, it clusters signals from product data into prioritized tasks that engineers can select from a to-do list, while a split-screen Command Center runs multiple agents in parallel. The system automates routine work, integrates seamlessly with PostHog for impact measurement and error debugging, and provides engineers the freedom to develop new features, with the goal of moving closer to full product development autonomy.

Publicly attacking competitors is often counterproductive, as it primarily resonates with existing fans while alienating potential customers and inadvertently promoting the competitor to a wider audience. Such actions open a company up to scrutiny, especially if they've made similar mistakes in the past, and can misalign with the collaborative ethos of open-source communities. Instead of direct attacks, companies should focus on humorous critiques of industry trends, provide honest product comparisons to build trust, utilize targeted advertising strategies like bidding on competitor keywords, and address factual inaccuracies directly with competitors. Ultimately, channeling competitive frustrations into improving one's own product is a more effective strategy for long-term success.

Building PostHog has been a dynamic journey marked by a successful pivot to finding product-market fit and rapidly growing its user base and revenue. The integration of AI has brought both challenges and opportunities, prompting a shift in focus towards AI-driven product and engineering work. The company initially launched an AI chat assistant, which faced issues but eventually improved through iterative development, leading to significant enhancements like a single query loop architecture and an AI-powered onboarding wizard. These advancements resulted in increased user satisfaction and retention, with AI becoming the default user experience. PostHog is now focused on making its platform more accessible to AI agents, transitioning from reactive to proactive AI functionalities, and developing long-running agents that align with user goals. This strategic shift aims to transform PostHog from a read-only analytics tool into a proactive solution-driven "doing company," enhancing productivity by automating low-level tasks. The company anticipates significant growth from upcoming innovations, even as their user experience undergoes temporary changes.

Real-time analytics platforms are essential for monitoring live user interactions on websites, offering features such as live dashboards, event streams, and user identification. This overview discusses several platforms, including PostHog, Google Analytics 4 (GA4), Plausible, Fathom, and Matomo, each with unique strengths and target audiences. PostHog stands out for its comprehensive developer tools and live user identification, while GA4 is noted for its broad integration within the Google ecosystem, despite privacy concerns and limited customization. Plausible and Fathom prioritize privacy and simplicity, providing lightweight, cookie-free solutions ideal for content and marketing sites, with Plausible offering self-hosting options. Matomo provides a feature-rich alternative, emphasizing full data ownership and compliance, making it suitable for enterprise or privacy-sensitive environments. The choice of platform depends on specific needs such as privacy, identified user tracking, integration capabilities, and the depth of real-time analytics required.

ClickHouse and DuckDB are two prominent OLAP systems, each with distinct features and use cases, leading to their adoption by companies like PostHog. ClickHouse is a highly configurable and scalable database designed for fast analytical queries on massive data distributed across multiple instances, making it ideal for real-time application analytics. In contrast, DuckDB is a lightweight, embeddable OLAP engine that excels at fast exploratory analytics on a single machine, suitable for smaller, spontaneous queries without the overhead of network communication. While ClickHouse is a dedicated server supporting horizontal scaling and complex queries with materialized views and indexes, DuckDB functions more like an embedded engine with flexibility in querying semi-structured data directly from files. PostHog utilizes both databases: ClickHouse powers their analytical engine for large-scale event data, while DuckDB, integrated with DuckLake and Duckgres, provides a manageable data warehouse solution compatible with Postgres-ready systems, enhancing their full data stack offering.