June 2025 Summaries

Building functional AI agents, distinct from traditional chatbots, involves understanding their core components: a model for thinking, tools for action, and structured instructions for behavior. AI agents are able to perceive input, plan, execute actions, and learn from interactions to achieve defined objectives, making them suitable for tasks requiring chained reasoning, handling ambiguous requests, or involving multiple steps. The process of creating these agents begins with a single, general-purpose agent and evolves into multiple specialized agents as complexity increases. Essential considerations include selecting appropriate models, designing clear and narrowly scoped tools, and providing comprehensive instructions to guide the agent's actions and responses. Security measures, such as authentication and safety protocols, are critical to prevent misuse, and testing is necessary to ensure agent reliability and resilience. The ultimate goal is to enhance productivity by automating repetitive tasks, allowing humans to focus on more complex problems.

AuthKit has launched CLI Authentication, incorporating the OAuth 2.0 Device Authorization Grant Flow for hosted AuthKit customers without additional cost. This feature addresses the challenge of securely authenticating command-line applications by allowing users to authenticate through a familiar flow similar to logging into streaming apps on TVs. Users run a command in their terminal, receive a code and URL, authenticate via their browser, and receive immediate authorization for their CLI tool. This process eliminates the need for storing long API keys or insecure credentials, enhancing security by avoiding long-lived credentials on disk and allowing users to revoke access through standard session management. It supports all existing AuthKit security features, such as multi-factor authentication and single sign-on. CLI Auth is automatically available for hosted AuthKit customers, with no additional configuration required, and detailed documentation is provided for implementation.

AuthKit Add-ons have been launched to provide seamless integration with tools like Google Analytics and Segment, offering enhanced insights into the user authentication journey for all AuthKit customers. These add-ons aim to illuminate the authentication process, enabling teams to track user interactions from initial sign-up through to successful login, thus providing valuable data for optimizing conversion rates and user experiences. By sending authentication events to analytics platforms in real-time, companies can gain comprehensive visibility into user behavior, identify friction points, and understand which authentication methods are most effective. The Google Analytics integration allows for detailed tracking of marketing attribution throughout the authentication flow, while the Segment integration facilitates the creation of unified user profiles by combining authentication events with product usage data. Prioritizing security, AuthKit ensures data protection by sending events server-side and prevents third-party script interference. Future expansions of AuthKit Add-ons will be guided by customer feedback, with potential integrations including RudderStack, Mixpanel, Amplitude, and others.

Supabase users now have the option to integrate WorkOS as a third-party authentication provider, enhancing authentication processes for both startups and enterprises by offering robust SSO and unified authentication without compromising developer experience. This integration allows for flexible, hybrid authentication flows, accommodating different user segments and supporting scalability of applications built on Supabase's PostgreSQL-based backend services. Users can configure WorkOS as the primary authentication provider, bypassing Supabase Auth, and align with existing customer tools without the need to rearchitect their stack. The integration is especially beneficial for businesses using Supabase alongside platforms like Railway, enabling secure, scalable SSO with minimal configuration. WorkOS provides enterprise-grade identity features and is free for the first million monthly active users, while Supabase charges a small fee per user when using third-party authentication.

WorkOS has introduced Feature Flags, a new addition to its platform, tailored specifically for B2B SaaS companies using AuthKit, allowing them to manage feature rollouts by organization directly within the WorkOS dashboard. Feature Flags enable developers to activate or deactivate specific functionalities at runtime without deploying new code, providing precise control over feature rollouts across customer bases and environments. This tool allows gradual feature rollouts, in-production testing, and swift deactivation of problematic features, addressing the more structured needs of B2B applications compared to consumer-focused products. WorkOS's Feature Flags are integrated into its platform, allowing seamless management of features, organizations, and authentication in one place, and are available for all AuthKit users without additional setup or infrastructure management.

The introduction of custom CSS support in the WorkOS AuthKit branding editor enhances the visual customization capabilities for teams, allowing them to apply their own CSS styles to AuthKit components while maintaining security and stability. Previously, AuthKit offered simple global branding options like colors and logos, which were suitable for quick and cohesive branding but insufficient for teams with complex visual guidelines. The new feature enables precise styling of individual UI elements, global styling across all AuthKit pages, and the use of remote images for backgrounds, ensuring a consistent authentication experience. Additionally, the update includes a targeting tool that simplifies element customization within the editor and a smarter back button that preserves unsaved changes, encouraging experimentation and making the editing process more intuitive and efficient.

MCP elicitation introduces a transformative capability in AI systems by allowing them to dynamically request additional context from users during runtime, thereby enhancing model performance and interaction. This new feature, introduced in the June 18, 2025, edition of the MCP specification, streamlines communication between AI models and clients by enabling structured dialogues that address missing information, such as user confirmation or authentication. Elicitation is crucial for resolving ambiguous scenarios where AI models require more context to function accurately, such as confirming actions, verifying identities, or clarifying user requests. The process involves an ElicitationRequest message from the MCP server to the client, which then prompts the user for the necessary information. While offering flexibility and adaptability, MCP elicitation also emphasizes security and user privacy, ensuring that no sensitive information is requested and that users can reject or cancel requests. This feature advances the development of context-aware AI systems, making them more modular, interactive, and capable of adapting to real-time user states without pre-defining every possible scenario.

Launching an Enterprise Plan is a critical step for products aiming to meet the high standards of large organizations, involving robust security measures, detailed administration, and scalable architecture. This process requires products to be enterprise-ready, meaning they must function reliably and securely within complex systems while providing a seamless user experience. Essential components include Single Sign-On (SSO) for centralized authentication, Domain Capture for appropriate workspace routing, and System for Cross-domain Identity Management (SCIM) for automated user management. Fine-Grained Authorization (FGA) and Role-Based Access Controls (RBAC) offer the necessary user access policies, while audit logs ensure compliance and security transparency. Products must also provide self-service admin tools and secure secret management to satisfy enterprise IT and security needs. WorkOS supports companies like Anthropic, Vanta, and Webflow by offering tools that simplify integration and enhance product readiness for enterprise demands, thereby facilitating the transition to enterprise-scale operations.

The integration of AI onboarding agents into SaaS products presents new challenges in identity and access management, as these agents require extensive access to tools and services to function effectively. Unlike traditional machine-to-machine authentication designed for predictable scenarios, AI agents operate more like digital employees, necessitating a rethink of security practices such as least privilege access, compliance, and authentication without a login page. This shift demands innovative approaches, including persona shadowing, delegation chains, capability-based tokens, and human-in-the-loop escalation to ensure secure and efficient operation. Emerging standards like UMA, GNAP, and OIDC-A are being developed to address these unique requirements, while industry leaders are adapting their systems to accommodate the growing presence of AI agents. As AI agents become more prevalent, potentially driving the majority of system interactions, the industry must collaborate to establish robust standards that maintain user trust and ensure secure digital futures.

AI startups aiming to capture enterprise customers must ensure their products are "Enterprise Ready," meaning they comply with the technical, security, and operational standards demanded by large organizations. This involves meeting stringent security requirements, integrating seamlessly with enterprise identity systems, and providing robust data governance and reliability guarantees. These enterprises prioritize trust, compliance, and integration over mere functionality and expect features like single sign-on (SSO), SCIM provisioning, audit logs, and high availability from AI solutions. WorkOS assists AI startups in bridging the gap between a functional AI model and an enterprise-ready product by offering solutions such as SSO, access control, and audit logs, enabling startups to meet enterprise IT requirements quickly and focus on their core AI innovation. By using WorkOS, AI startups can ensure their products are secure, reliable, and compliant, facilitating faster deal closures and fostering long-term trust with enterprise clients.

Engineers at a company identified inefficiencies in their API code, where the same SQL query was executed multiple times during a single HTTP request, leading to unnecessary database load. To address this, they implemented a request-scoped in-memory query cache within their database adapter, which would store query results during the lifecycle of an HTTP request, reducing redundant database queries. Using NestJS and TypeORM, they encountered initial challenges due to TypeORM's caching limitations and NestJS's request-scoping complexities. By leveraging nestjs-cls for continuation-local storage and extending TypeORM's repository behavior, they managed to maintain cache integrity without extensive code rewrites. This optimization resulted in a 30% reduction in database queries during testing, demonstrating a significant performance improvement without the need for a large-scale refactor, thus allowing the team to focus on product development amidst increasing service loads.

WorkOS has streamlined its SDK documentation, code examples, and changelogs by consolidating them under a single MCP server, which can be quickly installed via Cursor. This integration provides developer tools, MCP clients, and AI assistants with a unified interface to access WorkOS reference materials through four main tools: workos_search for full-text searches, workos_docs for fetching Markdown documents, workos_examples for retrieving code samples, and workos_changelogs for accessing version histories and release notes. Installation and setup are straightforward, allowing users to easily incorporate the server into any MCP client and enabling LLMs to utilize a progressive discovery workflow. This approach eliminates the need for navigating complex directory structures and allows for direct querying of the most current documentation, providing efficient and predictable access to WorkOS resources.

SCIM (System for Cross-domain Identity Management) is an open standard that enables seamless user identity management across different systems and apps. By automating the process of creating, updating, and deleting user accounts in multiple places, SCIM streamlines onboarding, reduces support overhead, and enables frictionless scale. Most successful SaaS companies have tapped into this quiet growth engine to fuel product adoption, drive expansion, and increase stickiness. SCIM is not just a backend protocol but a go-to-market tool that accelerates growth and unlocks scale, especially for productivity and collaboration tools targeting fast-moving teams. By integrating SCIM early in their journey, startups and scale-ups can offer security-conscious buyers confidence that their product can scale with their org, differentiate themselves from competitors, and start growing faster. With platforms like WorkOS making SCIM implementation radically easier, there's never been a better time to build it in early and reap the benefits later.

AI isn't magic, but rather a tool that can be used efficiently when properly managed. Professional knowledge workers use AI tools to accelerate their understanding and build context, which is then applied across multiple outputs. The key to productivity gains lies not in the tools themselves, but in knowing how to build and manage context. Context chaining is the practice of building deep understanding in collaboration with AI, then systematically applying that context across every deliverable needed. This approach allows professionals to direct AI to produce deliverables across multiple disciplines, maintaining coherence throughout, and becoming a conductor orchestrating AI capabilities rather than a user making isolated requests. By shifting the mental model from individual tasks to sustained context, teams can unlock extraordinary results and improve productivity.

AI systems are increasingly being designed with Human-in-the-Loop (HITL) approaches, which intentionally embed human oversight and judgment into the decision-making process. HITL systems recognize that AI can excel in efficiency and scalability but often struggle with nuance, context, or ethical complexity on its own. By incorporating human input at various stages of a system's operation, HITL aims to improve reliability, accountability, and alignment with human goals. This approach is particularly important in high-stakes domains like healthcare, finance, legal reasoning, or creative work where AI may not be able to replicate the level of nuance and judgment that humans can provide. HITL systems are designed to balance automation with oversight, ensuring that humans remain actively involved in the decision-making process, especially when outcomes matter most. As AI continues to evolve, HITL is becoming increasingly important for building trustworthy and aligned systems that respect human values and real-world constraints.

Connecting AI agents to systems is crucial for their full potential, enabling them to interact with external software ecosystems and become true collaborators. The Model Context Protocol (MCP) offers a standardized approach to system integration, providing a client-server architecture where AI agents act as clients and connect to MCP servers that expose system capabilities through a common interface. This abstracts system access behind a shared protocol, unlocking reusability, consistency, and security across the agent ecosystem. In addition to MCP, other building blocks for system awareness include REST APIs, direct database access, message queues, and agent orchestration frameworks. These strategies define what AI agents can do, from simple API calls to complex workflows involving multiple systems. Security and reliability are non-negotiable in production, requiring proper authentication, access control, circuit breakers, rate limiting, retry logic, and performance optimization techniques like connection pooling, smart caching, and parallel execution. As the ecosystem matures, expect greater standardization, more powerful frameworks, better observability tools, and a focus on design for flexibility, security, and modularity.

An organization switcher provides a smooth, intuitive way for users to change context and access the data and settings relevant to a specific organization. It ensures that each view, resource, and action is properly scoped to the selected organization, helping maintain both usability and security. To implement an organization switcher in a React application using WorkOS, you need to install the WorkOS React library, set secrets such as API keys and client IDs, and then add an organization switcher widget or manually implement authentication UI using the WorkOS API. The organization switcher allows users to switch between organizations without the friction of logging out and back in, ensuring a seamless and secure experience. With WorkOS handling SSO, user management, and organizational context, you can focus on implementing frontend and backend logic needed to switch organizations cleanly and securely.

Bearer tokens are compact strings that act like digital keys, granting access to systems and data once logged in. They're a core part of modern APIs and authentication flows, particularly in OAuth 2.0 protocols. However, bearer tokens are inherently risky as they don't verify who's using them, only that they're valid. Secure handling is essential to prevent risks such as token transmission over HTTP, insecure storage, leaking tokens in URLs and logs, long-lived tokens, missing or weak token validation, poor client-side token handling, overly broad token permissions, CSRF vulnerabilities, replay attack vulnerabilities, and insecure error messages. To use bearer tokens safely, always use HTTPS, store tokens securely, use short lifetimes, implement proper validation, monitor for anomalies, plan for compromise, minimize scope, implement token revocation/blacklisting, secure client implementations, log security events, and conduct regular security reviews to mitigate these risks and ensure application security.

Securing AI agents requires a comprehensive approach that addresses their unique characteristics and risk profile. From robust authentication and granular authorization to defending against both malicious actors and well-intentioned agents gone rogue, every aspect of your security architecture must be designed with AI agents in mind. WorkOS provides the enterprise-grade infrastructure that makes this comprehensive security approach practical and scalable.

The May updates for WorkOS include the introduction of MCP Authorization with AuthKit, which enables fine-grained authorization for agentic applications. Additionally, users can now update their email addresses via the Update User API or in the Dashboard, with changes triggering verification and unlinking of OAuth identities that no longer match the updated email. Custom attributes also support mapping from SAML responses during Single Sign-On (SSO), allowing fragmented identity provider configurations to be unified under a consistent schema. Furthermore, WorkOS FGA now supports two special context variables and a new helper function, making it easier to write dynamic, context-aware policies. The Admin Portal has also been improved with clearer guidance for domain verification, setup notifications, and easier debugging.