Agentic computing is here, as Microsoft and Nvidia recently announced. Nvidia’s RTX Spark technology is at the centre of this new way of thinking about your PC, and the company is aiming it squarely at creatives.
The idea is that local AI agents running on your machine can act as “teammates” rather than “tools”. RTX Spark lets you use them as productivity assistants, which could be a real help for anyone who struggles to get their admin done. But the bigger news is that the built-in agents (Hermes, Perplexity and OpenClaw) are designed to interact directly with a whole host of apps, including creative ones. You can read more in my RTX Spark explainer, which goes into the benefits of local AI as opposed to running it on the cloud and exactly what the technology can do.
The premise is that an agent saves you time on busy work, freeing you up to do more creating. Nvidia also hopes it will help specialist creatives become more multidisciplinary, a skill increasingly in demand. That raises a bigger question about whether this is a good thing. The more technology democratises skills, the more creatives are expected to do, and the implications for the creative industries are enormous. But for the individual creative who can use Blender but not DaVinci Resolve (for example), there is clear benefit, and it could help small teams punch above their weight.
I’m not a professional creative by any means: Photoshop is about the limit of my skills. But I wanted to see how these agents work with creative apps, so I picked Blender, which I don’t know how to use and which gave me the highest possible barrier to entry. A creative pro will explore multiple apps in more depth soon, but for now I tried some tasks in Blender to see how easy it was to dive in as a complete beginner, whether the agent could do the work of a teammate, how quickly it worked, and what the experience was like.
How I tested
- Laptop: ASUS ProArt P16 with RTX Spark, 32GB
- Agent: Hermes
- Model: Qwen3.8 27B running locally
- App: Blender.
The set up
Setup was straightforward, with instructions for installing the Hermes agent. Hermes then guided me through which software versions I needed, how to download them and how to grant the relevant permissions.
The Blender test
As a beginner, I started by asking the agent what it could actually do in Blender. It came back with an extensive list of abilities, from editing and scripting to rendering stills and setting up cameras and lights. It also said it could do “scene surgery”, such as moving an object from one scene to another, and batch file operations.
Sign up to Creative Bloq's daily newsletter, which brings you the latest news and inspiration from the worlds of art, design and technology.
Image 1 of 3
The barbershop edit
I ran a few tasks through Hermes, and it handled the simple ones really well. I started by asking it to open a stock file of a barbershop scene and make the lighting much warmer. It did this with a script instructing Blender, then offered to confirm visually that the scene looked warmer. I said yes, and it checked and confirmed. I then asked it to put the before and after images side by side and export, which it did easily.
Editing a head
Hermes struggled with more complex, multi-step edits on this first time round. I asked it to change the hair colour of a stock model with curly hair and add clown make-up to its face. It changed the hair and eye colour quickly and easily.
Image 1 of 6
The clown make-up was harder. Hermes spent a long time trying to render the additions before bringing back an approximation of the design. From then on it worked one step at a time, asking for approval almost every time, even for small iterations like bringing the mouth more in line with, well, the mouth. I stopped after about half an hour, and this is where we got to (above). If I had the skills and inclination, I could refine the model by hand, so it could work as a starting point for a more in-depth project. Alternately if I had more experience in Blender, I could probably have been clearer with my instructions, which may or may not have helped.
My mouse
Image 1 of 2
I then started a project from scratch and asked the agent to build a mouse with a long tail and big ears (see above). The result was a little rough looking (that mouth!), but it was built quickly and simply. The first render didn’t show the long tail because of the camera position, so Hermes asked if I wanted to change the camera angle so it could build a visible tail, which was helpful. I confirmed and it made the image quickly.
Moving one object to another scene
Next I asked it to remove a model from one file and put it in the barbershop chair in my barbershop scene. After about 25 minutes of prep work, it estimated the job would take 5-10 minutes, and promised me it could do it after I repeatedly checked it was sure. By my deadline it had been working for about an hour and a half and still hadn’t finished.
So the agent is slow when you first ask it to complete a new task. But the technology works by learning the skills you train it in, so when you return to a task it should know what to do, including logging any mistakes and feedback. If you understand the software well, you can probably speed this up with better prompts. With time invested, it’s easy to see how this could become more useful, whether for beginners or for repetitive tasks you’d rather avoid.
The experience of talking to the agent
The agent talks to itself constantly while working through more complex tasks. It’s unclear whether that is a genuine thought process or a simulation designed to make you feel in control. It is reassuring to see what it’s doing, but frustrating to watch it tie itself in knots over something that could be searched online in a couple of seconds. At times I wanted it to just say “I can’t do that” rather than go around in circles – but maybe it would have got there in the end if I'd had left it for long enough. That may be part of its learning process, and as it logs its steps into skill files the behaviour may lessen over time, but I can’t say for sure.
Final thoughts
As a beginner, I appreciated how Hermes offered a gateway into a creative app I had never used before. It lowered the barrier to entry in a practical way, which could help creatives who are branching out or new hobbyists. Human direction will still be needed to steer, choose and refine the final outputs, but for many people this could be a significant productivity boost. That said, it was slow and needed a lot of hand-holding on complex tasks, and the usual AI concern remains that the technology could replace human teammates. We’re a long way from anywhere near that skill that at the moment – currently it has potential as a useful assistant, which is exactly how Nvidia is pitching it.
It’s impressive how the agent communicates with Blender through Python scripts and keeps the user informed throughout. Creative software apps are reportedly being rewritten for Spark, which suggests the creative tech companies are behind it. If the agent keeps improving with training, it could become more useful for the tasks creatives would rather not spend time on, and for speeding up workflows. It's one to keep an eye on.
Stand by for our full testing overview from a creative professional across multiple apps, coming soon.
Georgia has worked on Creative Bloq since 2018, and has been the site's Editor since 2023. With a specialism in branding and design, Georgia is also Programme Director of CB's award scheme – the Brand Impact Awards. As well as immersing herself with the industry through attending events like Adobe Max and the D&AD Awards and steering the site's content streams, Georgia has an eye on new commercial opportunities and ensuring they reflect the needs and interests of creatives.