An iPhone and Apple Watch app described as eight levels of nets, product intent at the top and code at the bottom: how the Model Explorer, personas, the Interview, Goals and the new review layer keep a real application true while it is being built, and why the same mechanism fits any domain. …
Everything in this tour was generated by AI: a coding agent builds the nets through MCP, agent lanes and command lanes bring models into the process, and the net stays in charge. Shown on the real screens of a live agent software team and a website operations framework. …
Nine capability packs, 133 lanes, one model, built entirely by prompting Claude Code and Codex through the runtime’s MCP. A full run read back from its own event trail, and the claim it supports: a process built by an AI is debuggable in this runtime in a way an application is not. …
Give every service of a product a team of four persona nets, one product office above them, three rules the nets enforce, and let it develop the product. Agentic-Nets developing itself, with every screenshot taken live. …
Twelve live Studio screenshots and no hand-drawn diagrams: Hermann’s five nets as eight readable views, the web investigator’s crawl, analysis and reporting, and four more examples read straight from their places and lanes. …
The recipe behind Hermann: install Agentic-Nets (Desktop Lite or Docker Compose), connect Claude Code, and give six prompts in order, each ending with a proof on the live runtime. The prompts verbatim, the traps, the pictures, and what it cost. …
A net proposes the next iteration, waits for one human choice, writes the spec, hires a headless Claude Opus 5 inside that spec, measures the result against twelve factors, reviews its own diff and merges the pull request. How Hermann is built, what he shipped, and what is not true yet. …
Agentic-Nets is the first multi-layer agent runtime with a live application layer on top. Applications, humans, AI agents, deterministic transitions, local and remote executors, MCP/API and observability all operate on the same event-sourced Petri-net state while it runs. This article explains the architecture, the seven transition types, personas, governance, use cases and how the model differs from workflow automation and agent frameworks. …
n8n runs workflows. Agentic-Nets runs processes: live Petri nets where state persists, agents are governed, and the LLM cost curve is designed to bend to zero. Ten honest reasons, plus a section on where n8n still wins. …
Version 2.50.0 gives Agentic-Nets an MCP client. Agents can call any MCP server, but only the ones their inscription declares, with vault credentials and an audited trail. …
