Hermes Super Kanban pitch packages multi-agent project orchestration, but proof of autonomous delivery remains thin

Exact “Hermes Super Kanban Agent” version is not stated in the Reddit excerpt or transcript. Public setup material says Kanban requires Hermes v0.12+; release notes later describe durable multi-agent Kanban in v0.13.0. Access in the video is framed as an AI Profit Boardroom package: full zip file, walkthrough, prompts, and coaching, not a standalone independently reviewed release. The transcript claims a human approval gate, memory log/vault, gallery, live preview, and agent roles such as researcher, designer, and coder. Hermes Kanban Setup, 03:01

Documented Hermes Kanban is a dispatcher-based multi-agent work system, not merely a UI board. Official docs describe worker lanes with assignees, spawn mechanisms, per-task workspaces, environment variables, and lifecycle truth owned by the Kanban kernel: ready → running → blocked / done / archived. Workers report through kanban_* tools and reviewers gate completion. CLI docs describe per-board SQLite storage and many-board support. No public reproducible evaluation, benchmark suite, or ablation study verifies success rate, task quality, or autonomy claims for the “Super Kanban” workflow. Worker lanes, Kanban docs

• Vendor-reported: One sentence can become a plan, agent assignments, build, gallery item, and live preview after human approval. Evidence is the promotional video transcript, not an independent test. 00:55, 04:06
• Documented by project materials: Hermes Kanban supports multi-profile workers, task lifecycle, handoffs, heartbeats, and board tools. v0.13.0 release notes, Kanban docs
• Independent/third-party: Hermex review reports mobile access to self-hosted Hermes sessions, workspaces, dispatcher runs, and Kanban boards, but does not validate “Super Kanban” build quality. Hermex review
• No independent corroboration found: public sources do not provide benchmarked completion rates, defect rates, security posture for this packaged Agent OS, or reproducible demo artifacts.

The story fits a broader shift from chat-based agents to stateful orchestration: durable boards, task graphs, worker identities, memory, and human review gates. Hermes’ public docs position it as an autonomous agent with persistent memory, skills, tool gateway, messaging, MCP, and deployment options. The “Super Kanban” pitch appears to package those primitives for business workflows—especially SEO/content/app prototypes—rather than introduce a new foundation model. Hermes docs, Hermes Kanban Setup

The strongest limitation is evidentiary: the story is commentary plus a promotional transcript. Public docs support the architecture, not the claimed outcome quality. Safety concerns are concrete: CSA summarized an independent audit alleging severe default-configuration risks, and a GitHub issue reported Kanban/Docker behavior where a worker searched for kanban.db, consumed many tokens, and allegedly followed unrelated markdown content. These claims argue for sandboxing, scoped workspaces, approval gates, logging, and prompt-injection testing before business deployment. CSA note, GitHub issue #71456

For leaders, the practical opportunity is compressing messy handoffs: idea capture, task decomposition, assignment, progress visibility, and artifact storage in one workflow. For developers, the value is less “robot worker” than auditable orchestration around fallible agents. Treat it as an internal automation framework: start with low-risk repeatable workflows, require human approval before execution and release, instrument every worker action, isolate credentials, pin workspaces, and compare outputs against human baselines before using it in client-facing or production operations.