Thoughts on OpenClaw, Grasshopper and CAD Agents like Raven

Rhino AI
1

In this post I want to share some thoughts on how emergence made OpenClaw so successful and what that means for Grasshopper and Raven.

OpenClaw is the latest project creating significant buzz in the AI scene. For those who don’t know: OpenClaw is an AI Agent that lives on your computer and can do almost anything there (sounds like a great idea, right?). But here’s why people are loving it:

LLMs have been around for a while now - and they can be useful for some isolated tasks such as writing an email, but the promised ‘AI Asisstant for everything’ has not yet materialized to the extent we were hoping for. The dramatic rise in popularity of OpenClaw suggests that the project actually brought us a significant step closer to [whatever we were expecting].

Emergence

The reason why OpenClaw is so successful is that its Agent benefits from emergence. An LLM that can write emails is not that useful by itself. An LLM that can add events to your calendar is not that useful by itself either. But when you combine those simple skills, the Assistant can ‘schedule meetings with others’. Without adding, but from combining, you get a skill that is actually quite useful.

On Wikipedia ‘emergence’ is described like this:

emergence occurs when a complex entity has properties or behaviors that its parts do not have on their own, and emerge only when they interact in a wider whole.

Let’s consider how Grasshopper benefits from emergence and what that means for AI agents such as Raven. Grasshopper is well known and loved for its flexibility and customizability (you are probably aware of that that if you’re reading this forum). People use Grasshopper because it allows them to modify and tweak their workflows in ways that other CAD software doesn’t.

With around a thousand (!) plugins published on food4rhino, Grasshopper has been extended with capabilities that other platforms can only dream about. This is why Autodesk added Rhino.Inside for Revit. I mean, think about it: Autodesk allows you to install their competitor as a Plugin…

An AI agent that can model simple geometries is great. But what’s better is an agent that can do real-time physics simulations (Kangaroo), environmental analysis (Ladybug), load terrain data (Elk) or perform finite element analysis (FEA). And if you’re missing a feature, you can just add it yourself.

That’s why built Raven in Grasshopper - on its own Raven wouldn’t amount to much. But by integrating with the Grasshopper ecosystem early, Raven became the most versatile CAD Assistant out there right now.

I’m curious to hear what your thoughts are on this - where do you think we’re headed with CAD agents? Have you had success with automating workflows in CAD yet?

Max

2

How much of this was written with AI is usually the first question I ask. Tell-tale signs are the overuse of M-dashes, using words like “moreover” way too often, and marketing engagement speak to get people to respond. Sometimes I wonder if people are instructing their bots/agents to make human mistakes in their writing, so people believe its not AI slop.

That said, OpenClaw is a milestone we’ll be talking about for our lifetimes and beyond. Building out personalized autonomous agents for ourselves is going to be a must for anyone interested in keeping pace with emerging technolgies–and I group CAD/3D technologies in there, since most of us are in that camp. The question really is time. How much time do each of us have to invest in these technologies & major milestone platforms being developed? I’ve been putting in about 1 day per week, sometimes more, on Open WebUI and OpenClaw, and I’ll have to say, it’s starting to pay off. I’m not a coder by trade, more of a design technologist and professor of architecture, but it’s nice to be living out a life where the computational promise we saw in the late-'70s and '80s is truly coming to fruition. We’re not over-the-hump by any means, but something just happened with OpenClaw, 5.3 Codex, and Opus 4.6 that changes everything.

3

Hi Steve,

I’m disappointed (with myself) that you consider my post AI written. It is written entirely by me, not even proofread or translated through any tool.

But let me answer to what you said. I’m not sure if I would put ‘personalized autonomous agents’ and CAD agents in the same group. OpenClaw is clearly for ‘personal use’ if I may put it that way, while I see CAD agents more in a ‘professional’ setting. For two reasons: Firstly there are not that many CAD-personal-users (outside of this community). Secondly CAD agents are still way behind on capabilities.

But then again, you might be right, if we have enough time, we’ll get there at some point (but that also holds true for many things). And then you’ll finally be able to say ‘Hey [name of your Assistant’] - print me a new coffee mug with cool parametric patterns’ :grinning_face_with_smiling_eyes:

Max

4

Heya. I wasn’t making any accusations–I’m an AI optimist for the most part. However, I have grown accustomed to weeding through so much AI noise, lately, especially on LinkedIn. We are seeing an uptick here in Discourse, too. Maybe the use of LLMs is changing how humans format their own non-AI dialogues, too? To your opinion about there being a line between personal and professional AI use/outcomes, my own life is so much more hybrid, now. While not in a traditional or cutting-edge design office anymore, my classroom, work & strategies, and personal interests are intersecting much more than in the past 35+ years since the CAD revolution.

The coalescence feels like a brave new world.

5

Hey, first of all romantic labs did a good job. I just changed from win to Mac, because learning in my Bachelor Thesis on Revit (only Win Prop. what a mess?) I decided to go Mac and learning now rhino, a tool that’s cheaper than others, and quite more democratic and interoperable. But also learning grasshopper is quite hard now, again a program, and again an again. So the question must be for me, how can I create a design process for the future that works stable, and I hope those agents like Open Claw etc. will solve problems like "geometry” “materiality” “construction” in one. maybe we will step some day in a world that doesn’t rule on “ form follows function “ more “ form follows aesthetics” “ or form follows nature” as nature has a lot of algorithm inside, I am curious about all that, I think that ai stuff, could break the rules from all design history before, and bring complexity curves in simple language to our life. Greets from Germany

6

I think that we will never truly replace CAD workers because the integration between the different system, solutions and knowledges are still very very complex for AI to be able to handle, at least for now.

I also think that the agents will be another tool for good professionals that will lower the overall amount of avaliable jobs because of the boost on productivity of the professionals, kinda like it happened before and after auto-cad. Before you had one engineer to calculate, a projetist to design it, and a drawer/assistant to put it on paper. Now, at least in the places I’ve worked on, those 3 professions were merged into one, since it is all on the computer now.

However, everytime I tried to make something slightly complex, it failed, or because it was not able to truly make what I needed, or because it took so many prompt iterations that in the end was easier for me to have made it myself. Also, it sometimes feels like the AI agents are like a shotgun, you need to fix one tiny thing, and it ends up messing a lot of other stuff.

And, those scaling problems of having a lot of iterations or needing to babysit everystep of the project, are what are preventing me, as a freelancer, of using AI in my workflow. It fails to do the 100%, it usually does 70%, and the time it takes for me to understand the 70% it made to be able to finish the final 30% takes longer than if I made the other 70% myself.

As per smaller tasks, most of that don’t really need AI, just clever task generation, and that was already possible with the regular assistants we had.

IMO, the true development that made the difference is the MCPs not the agents itself.

7

I know that feel, it happens with all the AI applications all the time. And I agree that raven doesn’t master more complex tasks yet. We’re certainly working on that.

But it is still true that the capabilities that Raven has today emerge from the GH ecosystem, allowing beginner users to do things they would never be able to do without it.

Totally agree on that

8

sometimes autocad makes new macros based on my command usage history and that’s pretty nifty.

9

The thing is that, Raven can be a bit more organized on that, compared to regular programming

Maybe the use of clusters could help with that, separating stuff by what it is supose to do or something like that.
The visual side of visual programming can help Raven with that.

however, the need for conection to the raven cloud prevents me from using it at work, so I tried it only at home for small stuff.