Symposium: Exploring New AI Workflows (smallcultfollowing.com)

Symposium is an open-source experiment that stitches together an MCP server, a VSCode extension, and a macOS desktop app to create “taskspaces”: isolated clones of a project paired with an agentic CLI (the system currently targets Claude, VSCode, macOS and Rust). Each taskspace runs an agent connected to the Symposium MCP server so the agent can post logs, spawn new taskspaces, and call specialized MCP tools. Notable features include present_walkthrough (renders markdown walkthroughs with mermaid and anchored comments inside VSCode), an ide-operations tool for IDE queries, and Rust-focused helpers like get_rust_crate_source that check out crates and search examples to ground the model in real code. The author demonstrates fast context switching via taskspace-focused window management and shows how agentic code generation plus lightweight human oversight can accelerate learning new languages (the app itself is authored largely with model assistance in Swift). For the AI/ML community this matters because Symposium experiments with practical, composable workflows for agent-assisted development and collaboration rather than a single monolithic assistant. Technically, it leverages MCP for custom tool invocation, embeds references (<symposium-ref/>) to connect IDE selections back to agents, and envisions decentralization—crate-provided MCP guidance, WASM MCP servers, shared memory for bootstrapping contributors, and tighter issue/mentoring integrations. It’s an invitation to explore new agentic workflows, tooling primitives, and ecosystem conventions that could make model-driven development more reliable, reproducible, and collaborative.