June 2026 Summaries

Mike Cann compared Claude Code and Codex by having each build a React, Vite, and TypeScript collaborative project-management application using Convex, Supabase, and Firebase in isolated Docker environments. The six builds were evaluated for 40 functional requirements, security, and code quality, with additional runs measuring time and cost; manual functional testing was used after automated Playwright-based testing proved too unreliable. Claude Code generally produced more maintainable, modular, and better-typed code, receiving a code-quality grade one level above Codex across all backends, while Codex was typically two to three times faster and slightly cheaper overall. Both agents achieved full functionality with Convex and Supabase, though Firebase caused major difficulties: Claude Code exceeded the 45-minute limit and Codex required multiple repairs while still missing file-upload requirements. Security outcomes were broadly similar between the agents but varied substantially by backend, with Firebase earning D grades due to flawed authorization rules, Supabase receiving C grades for row-level-security weaknesses, and Convex receiving B grades because of unsafe file-serving behavior. Based on this benchmark, Convex appeared most suitable for AI-generated applications because its default transactions and reactive queries led to simpler front-end code and fewer severe authorization issues, although the author notes that results are limited to this application and methodology.

Convex 1.39 introduces a new feature that allows developers to define type-safe environment variables within their application's configuration, enhancing the reliability and clarity of code by using Convex validators to specify expected variable types. This development prevents runtime errors by ensuring that environment variables match the declared validators, offering immediate feedback and auto-completion in the code editor. While it currently supports only a subset of validators due to its implementation at the Rust layer, this feature represents a significant improvement in managing environment variables, mirroring the same code-centric approach that Convex applies to schemas and APIs. Although the feature doesn't replace the existing method using process.env, it provides a more robust alternative by generating a typed environment object that integrates seamlessly into backend functions, making it particularly beneficial for developers and agents who rely on precise type definitions. Additionally, components benefit from an upgrade that allows them to declare necessary environment variables, ensuring that any parent application using these components must provide the required variables, thereby preventing potential runtime crashes.

Alex Carrabre of Founders, Inc. discusses Mint, a platform that provides tooling, or a “harness,” for AI world models that turn text or images into navigable 3D environments usable in areas such as gaming, architecture, construction, visual effects, and event planning. Mint emerged after the team shifted away from crypto products, despite earlier applications reaching more than 100,000 users, because they believed the crypto label discouraged adoption; an on-chain room-customization project and early access to World Labs’ API led to the new product. After pivoting and adopting Convex as its backend, Mint reportedly grew from zero to 40,000 users in two months, and the team credits the platform’s speed and low latency while focusing next on improving its chat experience and expanding integrations, including a recently released MCP. Carrabre characterizes the venture as a bet that world models will eventually find product-market fit, though the leading application remains uncertain, and positions Mint as infrastructure for whichever use case succeeds. He emphasizes launching early and learning from real user behavior rather than delaying release in pursuit of a product founders assume is perfect.

Mike Cann introduces the Convex plugin for Claude Code, which can be installed through the Claude desktop app or CLI and is presented as more deeply integrated than the prior Codex integration. In demonstrations, Claude Code automatically recognized when Convex suited a requested collaborative family whiteboard application, asked clarifying questions, built and browser-tested the app in roughly 10–15 minutes, and delivered real-time drawing synchronization across devices without the user explicitly invoking Convex. A second test involved creating a Supabase-based pocket-money tracker, whose setup required manual project configuration and whose generated version did not synchronize updates in real time; Claude Code then migrated it to Convex, resolved an authentication error after feedback, and produced live synchronization between sessions. The plugin includes Convex-oriented design skills, connectors associated with the Convex MCP server, a Convex Expert agent, and hooks that appear to run type checking after code changes, although the presenter notes that Convex authentication remains less streamlined than Supabase or Firebase. Convex plans to continue expanding the plugin as Anthropic adds plugin capabilities.

The Convex Codex plugin for OpenAI's Codex coding agent enables developers to create and modify Convex backends directly within Codex, facilitating the development of real-time, multi-client applications without needing to manually wire up a backend. By invoking the plugin with the command @convex, users can scaffold new Convex applications or integrate Convex into existing projects, enhancing client-side applications with real-time syncing capabilities. This integration allows for seamless deployment with static hosting, providing shareable live URLs for collaborative apps. Although the plugin is new and requires explicit invocation, it demonstrates significant potential in simplifying backend development for AI-generated code by ensuring that the agent produces coherent, Convex-shaped code. The plugin's ability to handle complex features like live-syncing and shared cursors without detailed input from the developer highlights its effectiveness and suggests a promising future as it matures and expands its capabilities.

Convex migrations are necessary when schema changes would conflict with existing JSON documents, because Convex deploys schemas and application functions atomically and enforces explicit schema validation at runtime to support zero-downtime updates. The recommended approach is to first loosen the schema, such as making a new field optional or widening a field into a union, update functions to handle old and new data forms, run a batched data migration to backfill or transform existing documents, and then tighten the schema once all records conform. Some changes do not require migrations, including changes to empty tables, making schemas more permissive, or removing optional fields that have never been populated. Dropping fields is more complex than dropping SQL columns because Convex stores documents as JSON objects, potentially requiring scans and rewrites of all affected documents. Convex provides a migrations component for batched row processing and AI-oriented tooling to help automate the workflow, while failed migrations are generally easier to roll back before data changes begin and safer to roll forward after data has been transformed.