November 2026 Summaries

Mike Cann tests Expo’s pre-alpha Exact framework with Convex by building a real-time chat application that runs natively across the web, macOS, and iPhone from a single UI codebase written in Contract, Exact’s typed, Python-like frontend language. The demonstration shows messages, typing indicators, and heart reactions synchronizing through one Convex cloud deployment, while live theme-code changes are rapidly rebuilt and deployed to every device. Although Exact cannot yet run npm packages or provide the standard WebSocket and timer behavior required by Convex’s official client, Cann’s agent created a small custom client that communicates directly with Convex’s sync protocol. He highlights Exact’s potential for agentic development, particularly its controllable clock for more reliable animation testing, while noting that both the framework and the Convex integration remain experimental and that the protocol used is not a documented public API.

Mike Cann introduces Threadbear, a freely available MIT-licensed native macOS agentic development environment designed to run Claude Code and Codex while prioritizing speed, customizable workflows, and rapid human decision-making. The app organizes work into threads, automatically files them into projects through a “librarian” process, separates active work from an inbox of items requiring attention, and provides decision cards, comparison views, optional follow-up tasks, archiving, scheduling, and hibernation to reduce task-management overhead. Threadbear uses separate runners to execute agent work and Convex to synchronize lightweight thread state across desktop, web, and iPhone clients under the user’s own account, while keeping large logs local to control storage costs. Its “patches,” small TypeScript extensions stored within projects, let users add custom panels, actions, and project-specific tools, reflecting the creator’s goal of making the environment adaptable from within itself. Rather than distributing a fixed application, Threadbear is installed by forking its source repository and setting up a personal Convex deployment, allowing users to modify their copy while optionally merging future upstream updates with agent assistance.

Mike Cann introduces Batched Replay, a reusable Convex component that supports smooth, high-frequency shared interactions such as multiplayer cursors, collaborative drawing, and game ghosts without sending a database write for every client event. It collects rapid input samples into periodic batches, stores and distributes them through Convex, and replays them on other clients with their original timing, reducing writes while introducing configurable latency. Persisted batches can also enable reloadable drawings, time-lapse playback, and racing ghosts, while optional snapshots avoid replaying an entire history to reconstruct current state. Testing suggests the approach is most practical for relatively small rooms, roughly up to 50 active participants, because each update must be sent to all connected clients and function-call costs grow quickly with room size. Cann advises that it can be useful for collaborative tools and applications where historical playback matters, but may be unsuitable or costly for latency-sensitive, large-scale multiplayer games; the component is available through npm and the Convex components directory.