Short version first: the AI did its job. I forgot to set my Z height. If you only take one thing from this post, check your Z after any controller reboot.

Automate the mundane, keep the creative

Claude has an integration with Fusion 360, and I have been using it for the boring parts of CAM. It is not good at design decisions, and I do not want it making them. My rule is simple: maintain creativity, automate the mundane.

Why the spoil board needed new holes

The neck gets flipped mid job, so the blank has to register to the exact same spot both times. My original setup assumed a wider blank. The stock I can get locally is about five and a half inches wide, and I did not want an extra step cutting every board down to match. So I drilled new registration holes and put in new threaded inserts.

Give it real measurements and it gives you a real toolpath

The inserts had nowhere for their flange to sit. Instead of fixing that by hand, I grabbed the calipers and read Claude the numbers:

• Flange outside diameter: 11.78 mm
• Flange thickness: 1.4 mm
• Body diameter across the threads: 10.3 mm
• Between the threads, as close as I could measure: about 8.14 mm

It came back with a counterbore about 0.3 mm wider than the flange and 0.2 mm deeper than it is thick, so the insert sits just below flush. It also reminded me to zero on the same spoil board corner as before so the widened holes land on the old ones, and to pull the inserts out first so the end mill does not hit them. The recess came out clean and the inserts went in straight.

Garbage in, garbage out

Then it built the CAM package for a pine test neck: a numbered run order (attach holes, back roughing, the flip, then finishing passes), the right bits pulled from my Fusion tool library, and even a readme. That only worked because my actual tools were already entered in Fusion. If you hand an AI nothing real to work with, you are setting it up to make things up.

A lot of this neck design started with Martin over at Guitar Assembly. We are picking it back up with new stock dimensions and a settled truss rod model. Long term I want a spoke wheel truss rod because it is easier to adjust, but for this test it is a traditional two way rod. Somebody had already modeled both on GrabCAD, which saved me from modeling them myself.

Where it went sideways

The blank was 750 mm long. We hit a G-code error, Claude fixed the posted file, and I rebooted the controller. When in doubt, reboot. That reboot cost me my Z zero. I re-zeroed X and Y, rescued a wasp, and started the back roughing.

The lesson is not about AI. Any time the controller restarts, re-check Z before you hit run.

Tools and software from this session

Links below are Amazon affiliate links. As an Amazon Associate I earn from qualifying purchases, which costs you nothing extra and helps keep the shop running.

• Onefinity Woodworker X-50
• Makita RT0701C compact router
• Digital calipers, the most important tool in this post
• Threaded inserts for wood
• FastCap metric and imperial tape
• Fusion 360 and Claude

Follow along

One guy in a small shop on his father-in-law’s property in Lindsborg, Kansas. IT Director by day, maker by obsession. Equal parts problem solver, craftsman, and curious human. If you’re the kind of person who takes things apart to see how they work, or watches people who do, you’re in the right place.

Uncut shop sessions go up on YouTube and I stream on Twitch. Everything is @carverwoodworking. There is merch in the shop, or you can buy me a coffee.