https://automaton.run Software, hardware, and intelligence — notes on AI, robotics, fabrication, and building things that run. en-us Wed, 29 Jul 2026 04:30:13 GMT https://automaton.run/post/robotden-manifesto https://automaton.run/post/robotden-manifesto Robotics infrastructure keeps being rebuilt behind badge readers, and keeps dying there. RobotDen is a skunkworks run in public: move the missing middle into the commons and let it compound. Wed, 29 Jul 2026 00:00:00 GMT noreply@automaton.run (Bobby) Business https://automaton.run/post/teleop-customer-networks-pattern https://automaton.run/post/teleop-customer-networks-pattern Robots on networks you don't own: why VPNs and SSH escape hatches fail at scale, and the dial-out, sandboxed, audited pattern that doesn't. Tue, 07 Jul 2026 00:00:00 GMT noreply@automaton.run (Bobby) Robots https://automaton.run/post/implementation-companions https://automaton.run/post/implementation-companions Most founders write the book and stop. The next move — the implementation companion — is the one customers tell you to build. Here's what that looked like for Invisible Team Method and TemplateVault. Mon, 18 May 2026 00:00:00 GMT noreply@automaton.run (Bobby) Business https://automaton.run/post/mcp-won-tool-layer-unsafe https://automaton.run/post/mcp-won-tool-layer-unsafe MCP won the tool layer the way TCP won the network layer — by being everywhere, not by being safe. For agents touching real motors and real customer networks, what MCP is missing is exactly what production needs. Sat, 16 May 2026 00:00:00 GMT noreply@automaton.run (Bobby) AI https://automaton.run/post/quickbot-200-dollar-nav2 https://automaton.run/post/quickbot-200-dollar-nav2 A $1,200 robot to learn Nav2 is gatekeeping. The same robot exists for $200 used. Here's the platform-selection logic — and the open-source plan to take it the rest of the way. Wed, 13 May 2026 00:00:00 GMT noreply@automaton.run (Bobby) Robots https://automaton.run/post/80-20-agent-stack https://automaton.run/post/80-20-agent-stack Roughly 80% of the agent stack is commodity glue you should adopt. The 20% you build is what compounds. Here's how to tell which is which — and what robotics fleet operators have been getting right for a decade. Sat, 09 May 2026 00:00:00 GMT noreply@automaton.run (Bobby) AI https://automaton.run/post/home-lab-as-substrate https://automaton.run/post/home-lab-as-substrate Most home labs are a pile of boxes that each run one thing. The version that compounds runs everything — home automation, robotics dev, agents — as tenants on a single substrate. Here's what that substrate looks like. Wed, 06 May 2026 00:00:00 GMT noreply@automaton.run (Bobby) Hardware https://automaton.run/post/agent-frameworks-wrong-layer https://automaton.run/post/agent-frameworks-wrong-layer Agent frameworks are converging on the same primitives. The differentiation has moved a layer down — to the substrate that actually runs your robot. Here's what that layer looks like, and why robotics teams are already there. Mon, 04 May 2026 00:00:00 GMT noreply@automaton.run (Bobby) AI https://automaton.run/post/reaching-robots-behind-customer-firewalls https://automaton.run/post/reaching-robots-behind-customer-firewalls The last-mile networking problem that breaks ROS 2 deployments — and a new open-source reference architecture for fixing it. Ganglion, libp2p, WebAssembly, and lessons from cybersecurity. Sat, 25 Apr 2026 00:00:00 GMT noreply@automaton.run (Bobby) Robots https://automaton.run/post/dds-vs-zenoh-deployed-fleets https://automaton.run/post/dds-vs-zenoh-deployed-fleets ROS 2's default middleware was designed for a LAN. When your fleet leaves the warehouse, the assumptions break. A clear-eyed look at when Zenoh earns the switch. Mon, 20 Apr 2026 00:00:00 GMT noreply@automaton.run (Bobby) Robots https://automaton.run/post/building-agentic-workflows-with-n8n https://automaton.run/post/building-agentic-workflows-with-n8n Where dropping an LLM into an n8n workflow earns its keep, where it's dead weight, and the thing nearly every team gets wrong: error handling and reporting. From my book on workflow automation. Sun, 15 Mar 2026 00:00:00 GMT noreply@automaton.run (Bobby) AI https://automaton.run/post/jetson-orin-ros2-real-time-inference https://automaton.run/post/jetson-orin-ros2-real-time-inference A VLM on a $400 box will tell you when someone started and finished washing dishes at the sink. That's enough to rethink a lot of robotics telemetry. Tue, 10 Mar 2026 00:00:00 GMT noreply@automaton.run (Bobby) Robots https://automaton.run/post/esp32-mesh-network-sensor-array https://automaton.run/post/esp32-mesh-network-sensor-array Meshtastic is a beautiful piece of engineering. For my garden, it was the wrong tool. Here's what it's actually great for and what I switched to. Thu, 05 Mar 2026 00:00:00 GMT noreply@automaton.run (Bobby) Hardware https://automaton.run/post/code-generation-beyond-copilot https://automaton.run/post/code-generation-beyond-copilot Copilot is fine. What beats it for most teams isn't a fine-tuned model — it's stack-aware prompts, the right context, and an eval harness you actually run. Here's the stack. Sat, 28 Feb 2026 00:00:00 GMT noreply@automaton.run (Bobby) AI https://automaton.run/post/cnc-meets-generative-design https://automaton.run/post/cnc-meets-generative-design The generative-design industry sells topology-optimized billet parts. In 2023 I made a working replacement for my dishwasher in 15 minutes with OpenSCAD and ChatGPT. The interesting future lives at the cheap end of that gap. Fri, 20 Feb 2026 00:00:00 GMT noreply@automaton.run (Bobby) Fabrication https://automaton.run/post/startup-lessons-hardware-ai https://automaton.run/post/startup-lessons-hardware-ai Every funded robotics infrastructure company wants to be SaaS. The work is almost always consulting. Here's the honest version of the trade-offs from both sides. Sun, 15 Feb 2026 00:00:00 GMT noreply@automaton.run (Bobby) Business