This is done with ZERO CAD SDKs! I thought I had a demo before but nothing compared to this.
The teach pendant (Follow mouse → T → P) is online programming: you jog the robot, press teach, it records joint angles.
The edge-picking is offline programming (OLP) — specifically CAD-based OLP, and in welding it's usually called CAD-to-path.
How is this possible? The https://t.co/Qqz13P8BYQ MCP plugin is powering my native CAD tool. Both available on Github, check the links on the website.
3 HOURS OF MANUAL SETUP GOT AN AI BUILDING INSIDE REVIT. AUTODESK'S OWN VERSION STILL CAN'T.
Revit is the software behind most MEP work, the mechanical, electrical, and plumbing systems planned out digitally before a single pipe goes into a real building. Autodesk released Revit 2027 in April 2026, then followed it in June with an official MCP server, Model Context Protocol, letting outside AI tools like Claude plug directly into a live model instead of guessing from screenshots.
Claude Fable 5 connected without issue. It could read walls, ducts, pump schedules, dozens of parameters per object. It couldn't create a single wall, duct, or schedule, Tech Preview status, write access still locked on purpose.
A beginner MEP engineer wanted more: three pumps, a chiller connection, basic ductwork, built by description instead of menu clicks. The official server couldn't do it.
The fix was a community-built Revit MCP project instead, unofficial, 61 separate functions in total, including the ability to actually create and edit elements, the access Autodesk hasn't opened up yet. No official installer, no drag-and-drop, real technical setup before anything worked, close to 3 hours in this case.
Once it was live, a small MEP layout that would've taken most of a workday to build by hand was finished in 40 minutes.
Over in 3D art and design software, that gap doesn't exist. Blender's official MCP has had full read and write access since April 2026, backed by Anthropic as a funding partner of the Blender Foundation. Get a 3D scene wrong and you hit render again. Get a real pipe run wrong and you're filing a change order. That's the caution Autodesk is building into Revit.
We've all heard about CADSolver, the AI Native 3D Solid Modeling Kernel SDK/MCP Plugin that's revolutionizing CAD, but how do we get started with it?
Just install and say "evaluate my CAD" then give it some data. Nothing leaves your machine.
This turns the old months-long, NDA-heavy sales process kernel evaluation into a self serve process.
Built by https://t.co/Qqz13P8BYQ
https://t.co/Qqz13P8BYQ used our FeatureBoard MCP plugin to write a Text-To-CAD MCP plugin with our own AI Native BREP 3D Solid Modeling Kernel so that we can generate synthetic CAD test data to cover every situation in every operator that any customer has ever posted about in a public facing forum.
FeatureBoard turns product management into an incredibly high level abstract idea of how to get the data to describe successful products (make no mistakes!) without doing almost any work yourself.
Those tickets should take approximately five weeks to resolve at the current pace using subscription only compute.
In the words of Neo "Where we go from here is a choice I leave to you"
CADSolver Pricing for the AI Native SDK + MCP
https://t.co/atS85OPbkz
There's so much code flying out of this system that I cannot describe it to another person in a reasonable timeframe without breathlessly recounting two dozens features.
Leopold definitely over-levered but this is an incredibly well communicated investor letter...
He's also still up 80% this year which is better than most can say.
Hands up chefs! That's the end of July and Week 31!
How did you do? Let's take it to the judges table and see what they think.
https://t.co/Qqz13P8BYQ for the MCP Productivity tool that can handle CAD workflows with ease. I've got an AI Native CAD Modeler + MCP Text-To-CAD Plugin going as a solo project.
So now that the MCP Plugin is up to speed for CADSolver....
( my wife thinks the name needs work - leave a comment with a better idea~! )
I've had it start creating geometry for difficult conditions to probe capability, create a regression set, and then start targeting enterprise level functionality. At current pace I am estimating 3 weeks for our first customer's scoped operators and 6.6 weeks for the entire kernel to hit enterprise grade quality. We'll see I think the MCP thing might cut it down a bit to have really solid ways to tackle problems.
Get the plugin here https://t.co/GCld4UfhZT and start running CAD in Claude today with no other installs. Free for personal use, check out our pricing page for details.
https://t.co/atS85OPbkz
This is moving extremely fast so look for a lot of releases soon as this picks up steam. I have an evaluation concept cooking right now that I'll post separately that is just the bees knees.
Built by https://t.co/Qqz13P8BYQ
Both this week’s guests agree with Sam Altman, that we’re already in a version of the Singularity. But they can’t quite agree on why.
On TWiAI Ep 24, Grant Lee of Gamma and Philip Johnston of Starcloud help us define what that word actually means, and explain why - if the Singularity is here - things have not gotten more WEIRD.
0:00 Guest intros: Grant Lee (Gamma) and Philip Johnston (Starcloud)
1:22 Nvidia's open weight letter
2:16 Everyone loves open models
8:11 Starcloud's first H100 in space
19:47 Series A extension economics
20:39 How Gamma stayed lean and balanced work and fun
30:40 Betting on a compute-constrained world
35:48 Different approaches to pricing
44:59 Where does the AI moat really live?
57:41 Are we in the Singularity?
1:04:42 Starcloud's mission to Alpha Centauri
cc: @thisisgrantlee, @GammaApp, @PhilipJohnston, @Starcloud_, @jason, @alex
🎥 Watch the full episode here 👇
"We think about how human progress is predicated on communication. It's like, how do I get someone else to see what I see? This is a moment of extreme opportunity to actually drive human progress."
-- Gamma CEO Grant Lee
TWiAI Ep 24 is live now!
cc: @thisisgrantlee, @GammaApp, @jason, @alex
I had to text something rude back, the artificial turf cold text sale this morning touched a nerve. Fuck off!
For the non-phone issues, SlopRadar. Maybe coming to iMessages soon if I get many more of these?
https://t.co/x579vHa1Ad
Gentlepeoples of the world: Claude CAD does drawings!
It's a simple part, but between Claude CAD and build123d tools, we created a full, professional, machine layout drawing.
The drawing is fully parametric with the CAD code for producing the part. Claude fully understands ASME standard dimensions. We created the title block from scratch. SVG format.
The best part: No. More. Menus.
You want a thing, you don't need to go searching. No more combing through menus, trying to find the secret recipe for creating your hole callout. You just have Claude draw it. It's just coding.
This is combining the freedom of manual drafting with the parametric-ism CAD.
@yacineMTB That's a really great project!
I'm building a CAD Kernel with an MCP plugin for Claude, but my actual product is the MCP plugin I use to organize and structure the CAD work. You might find it useful since you're doing a lot of CAD heavy stuff.
https://t.co/GacF5XRU6R
anyways i'm designing a stereoscopic global shutter camera kit that can detect mosquitoes while hanging off your backpack. I have this running on a orange pi 6gb, i was able to write the camera driver for it its normally unsupported
I always like the Halo 3 online player count look.
Here's visits to https://t.co/GacF5XRU6R in my own custom analytics I rolled when I got the website online.
42 countries!
@TheStalwart Seems coherent enough.
Pay us rent for the technology through some mechanism that lets us cut in front of you at will and cut you off from it if your demand is higher than ours.
Awesome examples!
The LLM can just write so much more for the same concept than I would write. Heavy copy paste is going to show up in every hand coded repo!
If you've ever made music in a DAW it's so tempting to cut+paste+modify instead of write an original piece for an instrument that would be more interesting to hear.
bad news for the "agents write bad code" camp
Fable scored "how good" my 2 OSS projects were, using deterministic evaluators:
1) hunk (8k stars) - mostly codegen (2026)
2) counterscale (2k stars) - mostly handwritten (2024)
turns out I was mid