Could that bright project waiting in your notebook be live a week later with the right production line? In the old habit the Kanban board was only used for moving cards, ideas were forgotten and work stayed half done. In the new approach every idea left on the board is owned by a planning and building agent crew and turns real after a single approval. This piece explains the five-floor flow and its practical meaning in plain words.
In the old routine everyone drags cards, does the coding alone and lets bright ideas fade in a corner. The trouble is getting lost among tasks started with no plan and seeing who does what on one screen. A shared durable board makes tasks and owners visible and reduces clutter for the whole team. According to the Hermes docs published by NousResearch, the board is durable and tasks are shared across all profiles.
Why plan approval comes before everything
The heart of the new flow is showing the plan before work starts and getting plan approval for it. Under the AgentNative pattern, low-confidence requests are routed to approval and risk is caught early with care. With this step the user sees what will be built in advance and the wrong direction is fixed soon. The approval flow described in the OpenAI Agents SDK docs published on GitHub explains pause and resume steps and offers a similar control logic.
When an idea lands on the board it is first classified, then moved to the human approval zone and only approved work enters production. In production a manager splits the job while sub agents take parts like research and design. The layered setup may sound complex, yet the user only sees the result and the gallery. Compared with the visual board approach of the AgentKan project, this setup stresses more automation and review for teams. The visual board address is known as agentkan.ai and offers an open example for comparison.
The demos show a landing page built from a one-line request, a blog job split across research and design agents and an automation tool with several paths for a small business. Finished pieces like a meditation app and a habit tracker appear with previews in the gallery. These cases impress, yet they remain selected and polished scenes and deserve a calm reading. Even so the logic of the flow is clear and concrete enough to try.
On the technical side each task passes through a life cycle on the board, a dispatcher routes the work and the state is watched from the panel. Each worker acts with its own identity and moves cards forward with the tool set. This order builds one control point instead of scattered chat windows for the team. The Kanban document in the NousResearch repository on GitHub confirms the lifecycle through stages from ready to archived.
What the second brain and self review change
The memory side rests on writing each build record into daily notes and resuming from the same place the next day. When the user asks to fix the design of an older job the next day, the agent reads the log and makes the link. In the self review step the agent also checks its own output before handing over the work. These two layers work together for continuity and quality.
The memory habit may sound magical, but it is really steady note taking wired to automation for daily use. As daily records grow the agent recalls context, asks less and work moves faster with fewer repeats. Of course messy notes bring messy memory, so a plain template is a must for every user. A similar memory habit is already familiar to Obsidian users through the sync plugins listed in the community.obsidian.md directory.
The promo side is short, with the speaker naming his community and coaching work and leaving details aside. The closing message is clear, while most firms still drift across chat tools the early movers pull ahead. The call motivates, but decisions should still follow measured benefit and real trial results. A small pilot project is always wiser than a large commitment made too soon.
For starters the safest path is leaving one small idea on the board, watching it reach the product gallery stage and reading the plan at each step. Narrow the scope at the approval screen, keep memory notes plain and test the output with real users. This way inflated claims and true capacity part from each other in a fair test. Such a disciplined trial soon shows whether the system fits your needs.
Key moments
AI commentary
"The structure is genuinely useful for small teams because it replaces scattered trials with an approved plan and visible steps. The demo speed may still hide the share of templates and selected examples, so cautious curiosity is the right stance for now."
AI assessment
The strongest counter view is that the speed seen in such demos comes largely from ready templates and carefully picked cases, and real jobs with vague asks and messy inputs will not move at the same pace.
The gaps are also plain, error rates and cost figures are not shared, safety and access limits stay vague and the speaker promotion of his own community and coaching products creates a natural conflict of interest.
The practical takeaway for the reader is to run a small and measured pilot, taking one idea through an approved plan into production and noting time and quality as a roadmap that works for any team.
Sources
7 links; no other published story cites them. Stories sharing a link do not confirm each other; a source's origin is not inferred from how often it is cited.
hermes kanban · multi-agent system · plan approval · obsidian memory · automation