Official Luthiers Forum!

If the barrier to an automated dust collection system is the actual coding, this post might be of interest and encourage non-coders to tackle similar automation projects.

I had a chance to sit down with the Greenridge boys for creation and testing of their code using the Claude Code application running the Opus 4.7 model. While both gentlemen are heavily involved in AI projects for their respective employer/clients, neither considers themselves to be active applications developers/coders. With both having access to their locally-run AI models as well as what they refer to as foundation and frontier models, they chose to develop the shop automation with Anthropic's Claude Code application ($20/month for a month-to-month Pro subscription).

A brief summary of the project:

- 2 HP 2002-vintage Oneida cyclone with automated start/stop relay control
- 6" main ducting with two 6" branches, with the length of the controlling run 35'
- 8 connected machines (tablesaw, drum sander, jointer, bandsaw, router table, spindle sander, disk sander, and drill press
- 10 solenoid valve-controlled blast gates (one at each machine plus two branches)
- Arduino Mega R3 microcontroller
- Low voltage/12VDC 16x relay PCB
- 9 Hall Effect current switches (cyclone plus 8 machines) - essentially a SPST switch that operates based on the Hall Effect and is open when current is not flowing and closed at a user-adjustable current threshold.
- In automatic mode, the system triggers the dust collector and automatically controls all gates; a ten second run-out on the cyclone permits the sort of shuffle between machines for repetitive work without accumulating excessive cycles on the Leeson 2 HP cyclone motor
- In manual mode, the cyclone and 10 gates are directly controlled by the user. This allows use of the machines for metal work or for plastics, neither of which needs/should be used with dust collection if you are mulching or recycling. And a reminder: hot metal does not belong around wood dust.

Like most other AI instantiations, a terminal window may be used (think back to the pre-Windows days if a PC user or pre-Mac days if you used an Apple) or the context window. Using the context window is identical to the window which pops up on my browser and returns summaries of my searches or provides a button for more AI-generated info - this is the 'easy button' for those without a background in coding.

Mr. Stock had already written out a concise description of the system, desired function, and specific characteristics such as pre and post-operation runout time on the cyclone, etc. This description was placed in a desktop folder for AI use.

So from here, a step-by-step is illustrative of the level of effort needed to develop the code needed for the Arduino microcontroller:

1. I typed this message into the context window of Claude Code:

"Hi, Claude. I would like to have you develop application software for a Arduino microcontroller-run shop dust collection system. I have placed the system spec in <this folder>. and I grant you access to that specific folder for this project."

2. Claude reviewed the spec and asked for clarification as to the relay board (I should have provided the documentation in that same file folder as in step 1), but a quick answer (Active - High) allowed Claude to develop a plan of action which the AI then asked me to review. All looked good, so I directed Claude to move forward with coding.

3. In about 3 minutes, Claude dropped the code link into the context window, allowing me to open it in the Arduino Integrated Development Environment. (Arduino IDE). I did a code check for error (this is a push of an action button, so no expertise needed), then - receiving no error, I compiled the code and downloaded to an Arduino Mega R3 identical to the microcontroller already installed in the control cabinet down in the shop. Claude recommended a few optional changes (Master Kill switch for the dust collection system and optional coding in case the relay board was replaced at some time in the future with an 'Active-Low' versus 'Active-High' PCB), but other than loading the code from the IDE to the system Arduino Mega R3, install was just checking the code pin-out (e.g., which pin is connected to which relay/current sensor/manual switch, etc.).

The system powered up and operated per the spec, with just one tweak to code for run-out time in the manual mode (3 seconds is plenty, versus the 7 seconds originally used in the spec).

So in summary:

- If your hesitation with regard to tackling shop automation is coding the system as you wish it to operate, AI can handle that for your chosen microprocessor (Arduino, etc.) or mini-computer (Raspberry Pi, etc.)

- Cost to develop AI code is $20 or less for small shop systems (a 30 day subscription at the lowest or at mid-level), with most microprocessor development environments being no-cost applications

- AI will walk you through the process, explaining as would a good teacher if asked to do so (aka 'explainer' mode) and asking questions to clarify any issues with your system specification

Editorial:

I use AI scribes in my telemedicine practice, but have never been terribly impressed. Claude is a different animal entirely, anticipating further requests and conducting the same sort of conversation during the development effort as one might have with a motivated, eager junior colleague. I had the sense that we had just scratched the surface as to capabilities, and that the application (Opus 4.7 model) was just a bit beyond an idle for the task requested.

If those that found this post useful want me to post the system spec and the developed code, I'll be happy to do that. AFAIK, most of the 3D model files for the gate adaptors and gate filler plates (needed for the ProTec 6" gate to tighten things up) are posted on Thingaverse. There are also a few treads Mr. Stock posted over on Sawmill Creek if photos of the gates or control hub would be of use. I don't pretend to have good answers for his system choices or detailed gate or controller box layout, so best to pursue answers over on that site versus a relay effort here.

Right turn, Claude.

Wow

June 05, 2026:
Anthropic says Claude writes 80% of its own code and the world needs a plan to hit the brakes"

https://thenextweb.com/news/anthropic-c ... opilot.com

Kind of spooky....

_________________
Measure Twice,

Karl Borum

So-called 'lights out' code factories are where we are headed, according to the various AI gurus I surveil on YouTube. The only real job for the humans running the effort is careful initial description and application of constraints, and even this role disappears once the first versions of the code is produced. If you are wondering how all of these new apps are being developed and heavily marketed in a relatively short period of time, I suspect AI is doing the lion's share of the work and advertising budgets are reflecting the cost reductions in production.