March 2026 Summaries

FLORA, a company focused on visual iteration, has enhanced its creative workflow with FAUNA, an AI-powered agent designed to simplify and enrich the creative process by acting as a design partner. Originally, FLORA's node-based creative workflow canvas required users to think like workflow designers, which often interrupted the creative process. FAUNA removes this burden by allowing designers to start with concepts, automatically generating and iterating on visual content. Built on the Vercel AI Stack, FAUNA utilizes an integrated system that combines AI SDK and DurableAgent for seamless orchestration and parallel processing, enabling efficient visual ideation. This integration not only supports FLORA's goal of facilitating non-UI/UX design work but also frees up the team to focus on product development rather than infrastructure challenges, thereby fostering a more profound understanding of user needs and enhancing the overall creative flow.

The text highlights the increasing reliance on coding agents in software development, which generate code at unparalleled speeds but can also introduce significant risks if not properly managed. It emphasizes that while these agents can boost productivity, they often create a false sense of security, as their output may pass superficial checks but remain blind to production-specific realities and risks. The distinction between leveraging and relying on AI is crucial; leveraging involves using AI to enhance speed while maintaining full ownership and understanding of the code and its implications. The document advocates for building robust infrastructures that inherently ensure safety by design, such as self-driving deployments and continuous validation, where problems are contained and rectified automatically. It stresses the importance of executable guardrails and ongoing system stress tests to ensure resilience. The overarching message is to leverage AI's capabilities while retaining strict judgment over what code is shipped, ensuring that engineers are comfortable owning the risks associated with their changes.

The Custom Reporting API for AI Gateway, now available in beta for Pro and Enterprise plans, addresses the challenges of fragmented usage data across different providers and keys by offering programmatic access to comprehensive cost and usage metrics, including BYOK (Bring Your Own Key) requests. This API enables users to track expenditures by model, provider, user ID, custom tag, or credential type, thus facilitating a detailed analysis of costs per feature, customer, and pricing tier from a single endpoint. An AI platform with over 200,000 users previously used a separate proxy layer for cost tracking, but during the private beta of the Custom Reporting API, they consolidated their tracking efforts, saving over $80,000. The API supports tagging requests with customer-specific identifiers, allowing for accurate cost attribution and proactive usage monitoring. This system captures data regardless of whether users bring their own API keys or use AI Gateway credits, offering insights into per-customer and per-feature costs, internal usage across models, and assisting in budgeting and pricing decisions based on actual unit economics.

SERHANT., an AI-native real estate and media company founded by Ryan Serhant, utilizes its proprietary AI platform S.MPLE to enhance efficiency and productivity for real estate agents by automating repetitive tasks like market analysis and contact management. Under the leadership of Jeremy Bunting, VP of Engineering, the company successfully scaled from an internal pilot to accommodate over 900 agents without replatforming, leveraging technologies such as Next.js on Vercel and a React Native iOS app. The team prioritizes flexibility and rapid iteration in an evolving AI landscape by using Vercel's AI SDK, which allows integration with multiple models such as OpenAI, Claude, and Gemini, optimizing for cost, speed, and complexity. By maintaining an adaptable infrastructure, SERHANT. keeps its focus on innovation rather than constant rebuilds, ensuring its agents have a competitive edge. The company also boasts a strong media presence with SERHANT. Studios and educational outreach through SellIt.com, further solidifying its position as a leader in the real estate industry.

v0 and new.website have merged to enhance their shared vision of enabling users to easily create complete, production-ready software with AI. Originally, new.website aimed to simplify the creation of aesthetically pleasing websites by incorporating essential tools such as built-in forms, SEO features, and content management, which minimized the need for users to engage in complex coding or configuration. This integration with v0 focuses on leveraging new.website’s strengths in built-in primitives and design to develop agent capabilities, which will enhance core functionalities like forms, content, and data processing, making them more intuitive and agent-aware. This collaboration is set to streamline the process of building and deploying functional software, aligning with v0's mission of delivering high-quality, user-friendly web solutions.

Turborepo achieved significant performance improvements, up to 91% faster execution, by optimizing the computation of its task graph in large repositories, with efforts led by Anthony Shew. This was accomplished through a combination of AI coding agents, traditional engineering practices, and profiling tools, such as Vercel Sandboxes and Markdown profiles, which facilitated efficient performance analysis and optimization. Key optimizations included parallelizing sequential operations, reducing redundant allocations and syscalls, and leveraging stack allocation over heap allocation, which led to substantial reductions in execution time and variance. Despite the use of AI agents, the process required active human oversight for decision-making and verification, demonstrating a successful integration of AI support in software engineering while maintaining control over strategic directions.

In an effort to streamline workflow augmentation through chatbots, a company developed AI agents that initially faced integration challenges across various platforms such as Slack, Discord, GitHub, and Linear. This led to the creation of Chat SDK, a TypeScript library that allows developers to build bots operable across multiple platforms from a single codebase, using adapters to handle platform-specific behaviors. The tool abstracts messaging API quirks, enabling seamless deployment and interaction across platforms like Microsoft Teams, WhatsApp, Telegram, and Google Chat. Chat SDK also facilitates real-time AI streaming and offers pluggable state management with support for Redis and PostgreSQL. As a significant advancement, it extends its "write once, deploy everywhere" model to WhatsApp, accommodating messages, reactions, and multimedia. Open-source and in public beta, Chat SDK aims to unify and simplify the deployment of chatbots, offering documentation and best practices to aid developers in creating efficient multi-platform bots.

The Knowledge Agent Template is an open-source, file-system-based agent designed to simplify the deployment and customization of knowledge agents by eliminating the need for vector databases, embedding models, and chunking pipelines. Built on Vercel Sandbox, AI SDK, and Chat SDK, it allows for the integration of multiple data sources and platforms like GitHub, YouTube, and Discord, using simple filesystem commands such as grep, find, and cat for deterministic and explainable search results. This innovative approach leverages large language models' inherent understanding of filesystems to improve efficiency and reduce costs, offering a more transparent and debuggable alternative to traditional vector-based systems. The template supports deep integration with various AI SDK models, automated cost optimization based on question complexity, and includes built-in admin tools for comprehensive management and troubleshooting.

AI-native startups like Relevance AI and Leonardo.AI are leveraging operational models that bypass the need for dedicated infrastructure teams, allowing them to scale efficiently and focus on product development. Relevance AI automates workflows across platforms such as Salesforce and Slack, while Leonardo.AI processes millions of images daily for creative professionals, both achieving global scale without manual intervention. This approach is thriving in the APAC region, where AI venture capital is abundant but DevOps talent is scarce, enabling startups to allocate resources towards innovation rather than infrastructure management. Utilizing platforms like Vercel, these companies have dramatically reduced build and load times, enhancing their ability to innovate rapidly and meet the demands of expanding user bases. As the AI startup ecosystem in APAC grows, the success of these lean operational models suggests a shift towards prioritizing product development over expanding operations teams.

Durable, an AI business builder, simplifies entrepreneurship by automating tasks such as website creation, SEO, and customer relationship management, allowing entrepreneurs to focus on their customers. Initially aiming to reduce the friction faced by small business owners, the company transitioned to a multi-tenant platform run on Vercel to handle millions of customer sites with varying traffic patterns and operational needs. This shift enabled Durable to manage complex AI workloads efficiently, ensuring tenant isolation and cost-effective scalability. By consolidating their infrastructure and leveraging Vercel's capabilities, Durable's small team of six engineers has achieved significant operational efficiency, allowing them to serve over 3 million businesses and process 1.1 billion tokens daily without a dedicated infrastructure team. As a result, each team member can deliver output ten times greater than before, reflecting a shift towards a more agile and scalable business model that could define the future of tech operations.

In April, Vercel launched its Open Source Program to provide maintainers with resources, credits, and support to enhance their project development capabilities, and the winter 2026 cohort has recently been welcomed. This cohort includes a diverse range of projects spanning from AI-native applications and developer infrastructure to unique design systems and creative tooling. Among the notable projects are Answer Overflow, which enhances accessibility of Discord server content for search engines and AI; UI TripleD, offering comprehensive UI components; and hot-updater, a React Native OTA update solution. Other projects, such as Screenshot Studio and Takumi JSX → Image, focus on simplifying design and image generation tasks, while initiatives like APIs.guru and YourDigitalRights.org aim to streamline API usage and online privacy management, respectively. The program highlights Vercel's commitment to supporting open-source innovation and its growing community of creators, encouraging further participation and development through upcoming application opportunities.

The Vercel AI Accelerator has returned, selecting 39 early-stage AI startups from across the US, Europe, Asia, and Latin America to participate in a six-week program designed to foster innovation and growth. These teams, representing diverse industries, gain access to Vercel's infrastructure, significant partner credits, mentorship from VC experts, and weekly sessions with industry leaders. The 2026 cohort includes startups such as Alloy Robotics, Archival, Ares, Carbyn AI, and many more, each offering unique AI-driven solutions ranging from robotics to security and healthcare. The program culminates in a Demo Day, where teams present their developments to AI leaders, venture capitalists, and other stakeholders, with top performers receiving substantial credits, prizes, and potential investment from Vercel Ventures. This initiative has previously helped participants secure venture funding and enterprise partnerships, and it continues to be a launchpad for AI innovation.

Notion Workers allows developers to write and deploy code that extends the functionality of Notion by integrating external data, automating tasks, and calling APIs. These Workers operate on the Vercel Sandbox, which provides strong isolation through ephemeral Firecracker microVMs that ensure security by preventing unauthorized data access and credential exposure. This setup supports various use cases, such as syncing CRM data, triggering automations, and enhancing AI agent capabilities, while maintaining safety and performance. By leveraging Vercel Sandbox, Notion can focus on enhancing developer experiences without the burden of building complex infrastructure, as it aims to transform itself into a robust developer platform. Vercel Sandbox offers essential features like credential injection, dynamic network policies, and filesystem-based snapshots, making it an ideal choice for handling untrusted code execution at scale.

Vercel's CDN offers a flexible solution for enhancing security and functionality across various applications, such as Discourse or WordPress, without requiring a full migration to the Vercel platform. By using Vercel's CDN, users can benefit from features like firewall protection, DDoS mitigation, Web Analytics, and Bot Management, which help protect applications and provide valuable insights. The Vercel Community exemplifies this setup by proxying a Discourse application through Vercel's CDN, while certain components, like live sessions, run directly on Vercel using Next.js. Vercel's microfrontends allow for incremental modernization of applications by mounting Next.js apps on the same domain, enabling the creation of new pages or customization of existing ones without extensive changes to the core application. This approach allows for a seamless blend of legacy and modern solutions, offering a secure and scalable path forward without risking a complete platform overhaul.

Commerce application development has significantly evolved, focusing on quick iteration, global deployment, and leveraging AI tools for creating storefronts, checkout flows, and subscription logic. The integration of tools like Vercel CLI and Marketplace with Stripe has made the process more streamlined, eliminating manual steps and reducing friction in moving from test to live environments. These integrations allow developers to create and deploy commerce applications more efficiently, with Stripe accounts being connected directly to projects, automating environment variable management, and ensuring secure key exchanges. The collaboration between Vercel and Stripe introduces a single integration setup for accepting live payments, enhancing both developer experience and security by automating key management and reducing human error. This new approach allows for seamless transition from application development to generating live revenue, integrating payment processes directly into the development workflow.

Handling redirects at scale poses challenges related to latency and system costs, especially when managing millions of redirects. Vercel previously managed redirects via routing rules and middleware, which became inefficient as the number of redirects increased. To address this, a new approach was developed using Bloom filters and JSONL file format to manage redirect data. Bloom filters, which can quickly determine if a redirect might exist, reduce unnecessary processing by skipping the JSON parsing for most requests. To further optimize, the redirect data is sharded, allowing for smaller data slices to be processed only when necessary. This approach significantly lowers the memory usage and latency for redirect lookups, particularly by employing binary search over sorted keys instead of parsing entire JSON shards. This design reduces per-request CPU demands and eliminates latency spikes, making redirect handling faster and more efficient. Bulk redirects are now available for Pro and Enterprise customers, allowing for large-scale migrations and management of broken links with a capacity of up to 1 million redirects per project.

Slack Agent Skill streamlines the process of creating and deploying Slack agents by leveraging coding agents like Claude Code, OpenCode, Cursor, and GitHub Copilot to coordinate the necessary systems and automate steps. The skill uses a wizard to guide users through the entire process, from setting up a project with the Slack Agent Template to deploying on Vercel, ensuring each stage is handled efficiently. This includes configuring OAuth scopes, event subscriptions, and slash commands, setting up environment variables, and deploying to production. The agent can perform tasks like holding multi-turn conversations, reading channels, and executing custom tools for specific actions. Workflow DevKit enhances the agent's durability by allowing it to pause for human approval and resume conversations, while AI Gateway provides access to various AI models. Vercel Academy offers a comprehensive course to further extend and understand the capabilities of Slack agents, ensuring users can effectively manage the lifecycle and deployment of their agents.

Avalara has significantly transformed its approach to product development and customer interaction by integrating Vercel's AI-native workflow tool, v0, which enables rapid prototyping from plain language descriptions. This shift has allowed Avalara to quickly develop two patent-pending products and streamline its design and production processes, reducing the need for extensive documentation and long handoffs between departments. By leveraging v0, Avalara addresses the challenge of supporting customers across a vast array of systems by creating a Chrome extension that integrates seamlessly into user workflows, providing real-time assistance and visibility. The new approach fosters direct communication and collaboration, accelerating the company's ability to innovate and align product design with strategic goals, ultimately reflecting a broader cultural shift towards iterative building and immediate feedback within the organization.

PEP 827 introduces a proposal to enhance Python's type system, making it more expressive and composable, similar to the programmable types in TypeScript but tailored to Python's runtime model. The proposal aims to provide standard type-level building blocks for introspecting and constructing types, facilitating both type checkers and runtime tooling. This initiative seeks to align Python's typing capabilities with its dynamic nature, ensuring that constructs like metaprogramming and decorators remain compatible with static typing without necessitating typechecker plugins or excessive boilerplate. Highlighted by examples such as the implementation of utility types like Pick and Omit, the effort focuses on giving Python a programmable typing core that retains introspectability at runtime. This development is crucial for enhancing the expressiveness and thoroughness of type checkers and frameworks, enabling more reviewable and succinct code, even as automated agents increasingly contribute to code generation. The ongoing debate and revision of PEPs underscore the commitment to improving both Python and TypeScript ecosystems, emphasizing the enduring importance of programming language syntax and tooling.

Waldium, co-founded by Amrutha Gujjar and Shivam, is an agentic CMS and blog hosting platform that revolutionizes content creation and distribution by automating research and providing dedicated Model Context Protocol (MCP) endpoints for every customer, enabling their content to be directly accessible within AI assistants. Initially frustrated by infrastructure complexities, Amrutha and Shivam found a solution in Vercel Platforms, which allows Waldium to manage hundreds of customer blogs efficiently from a single Next.js application. This infrastructure enables seamless deployment and dynamic routing, significantly reducing overhead and allowing the team to focus on innovation. Waldium's approach not only streamlines content distribution but also transforms it into a queryable resource, as evidenced by clients like Sapra AI, who drastically reduced research timelines and increased content output. This model represents a new standard for businesses, emphasizing the importance of integrating data into AI workflows and highlighting the potential of well-thought-out infrastructure design.