Spatial intelligence in AI is on the cusp. After language, LLMs and then pixels, AI research is about to explode into the 3D world. This video is highly theoretical and granular. These CalTech scientists really get going on 3D applications of spatial intelligence at the 19 minute mark.
interesting explanation of using 2D pixels to create 3D models. they specifically state the intention of making AR and game environments more affordable.
the previous video is similar to other design engineering software but what I conclude is that any well-defined process with quantifiable results can be automated.
so what about us? for ID this might include Ergonomics to start with but much of human factors over time.
Also, Sustainability could be included perhaps performing an LCA from a BOM for example.
even if AI fulfills all these functions, they will be good at the basics but incomplete and need human experts to hammer out corner cases, and they will be separate functions needing human expertise to fully integrate into a final product.
My question is if AI can do all the grunt work how can an entry-level employee gain the expertise specific to that industry or product line?
Zhu definitely helps put things into perspective. There’s also the part where we can’t actually predict how fast AI will become practical.
Hi everyone,
I’ve been following the discussions here, and there are so many insightful conversations around AI’s role in supporting creative workflows—I would love to pick your brains more on this topic! 
Also, apologies for the bit of solicitation here, but I thought this could be a great place to connect with like-minded folks who are passionate about AI and design.
I’m a Postdoctoral Researcher at the Human-Computer Interaction Institute at Carnegie Mellon University. Our research team, including Dr. Aniket Kittur and Dr. Nik Martelaro, is working on AI-based computational tools to support designers, engineers, and R&D scientists in product innovation.
We are particularly focused on enhancing creative processes by systematically exposing professionals to cross-domain inspirations—moving beyond serendipity to structured discovery.
For example, the Odon device, a groundbreaking solution for assisting difficult childbirths, was inspired by a parlor trick for removing a cork from a wine bottle. Similarly, NASA has drawn inspiration from origami for foldable solar panels, and the Wright brothers modeled aircraft control on bicycles. We want to scale this process and help innovators uncover transformative ideas more systematically.
Some of our past research includes tools like BioSpark, which enables analogical transfer from nature using AI, and Inkspire, which supports AI-assisted sketching for designers.
We’re looking for designers, engineers, and product innovators to collaborate with us! As an advisory board member, you will:
If you are interested, drop me a message, and I’ll share more details on how to get involved. Also, if you know others who might be interested, feel free to spread the word!
Looking forward to connecting and hearing your thoughts!
Thanks,
Vikram
Typing to your CAD program seems better than hunting for seldom-used tools in drop-down menus.
Going by the example in that video, I’m not convinced.
That seemed to take 5 times longer to do.
hide / show / select / move
Having every element and layer naturally named helps immensely when tailoring for Large Language Model demos.
I compared Vizcom, Gemini and Adobe Firefly for making renders and iterating on concepts.
Firefly was garbage in my tests. Maybe there is some key to getting it to work, but I couldn’t get anything to work beyond the photoshop background erase and stuff like that.
Gemini is actually OK for renders, but it hallucinates features a bit. It loves to add USB-C to every product you ask it to render. Also, the text based interface makes it clumsy. For iterating, it was garbage. It just had no idea how to make another version of whatever I was trying to design.
Vizcom works really well for renders. I have to say, I think it’s a Keyshot killer. I tried pasting a screenshot of Solidworks into and making a render and it was spot on after 1-3 interations. I also tried very crude sketches and it generally understands what my intention is. It also doesn’t add useless stuff like Gemini. The visual interface and ability to make multiple renders at the same time makes it really quick and easy to use. However, the iteration is pretty basic. I tried and tried and maybe I get one OK concept for every 10 variants I tried. It might be easier if I was doing color studies, but I didn’t try that, I was looking at forms and details.
I’ve been using Vizcom with my students, and have been impressed, for two unexpected reasons.
the workbench encourages structured explorations and not the typical AI “try stuff and forget” process
It emphasizes that drawing is thinking. Sure it will make up a product image from a line, but it really demonstrates the classic power of sketching, Using a iPad / Wacom is a game changer.
It’s not CAD, it doesn’t really do design, but it could be useful at the end of sketching process. somehow it renders faster than a render. change materials? no problem.
I agree with Dan’s comments having been in ID since the early 70s. What I’ve been struggling with and ultimately what led me to create PA-AI is the realization that machines, LLMs, and AI can generate incredible amounts of information, but they still don’t truly understand what it means to be human. They may even be a great design tool, that I don’t know as we believe the hand mind connection in design remains essential to the Human experience.
LLMs don’t understand our emotional needs, aspirations, fears, anticipation, dreams, memories, or the deeply personal experiences that shape why we connect with certain products, brands, services, and ideas over others.
That’s the gap.
I kept coming back to one simple thought:
If LLMs are the engine, then the Human Intelligence Layer is the navigator.
That’s exactly what we’re launching at PA-AI . Its an overlay on top of AI and LLMs grounded in 45 years of human-centered innovation at RKS Design and 33 years of Psycho-Aesthetics®.
My belief is that the future of AI won’t belong to systems that simply generate more content faster. It will belong to systems that better understand people — what they value, what they trust, what inspires them, and ultimately what they adopt and love.
To me, this is where design doesn’t just participate in AI — it rises above the mix and helps guide it.
We’ve opened a waiting list and are offering free access for design students because I believe the next generation of designers and innovators will help humanize AI in ways the world desperately needs right now.
Take a look, sign up for the waiting list, we just weeks away from launching: https://www.pa-ai.ai
Back in the day, Arizona had a bill coming up for a public vote regarding whether or not they would continue to permit usurious payday lenders to continue to operate in the state. The radio ad I remember most was one talking about important it was to vote yes and “show the payday lending industry that they would have to abide by rules” and other such statements. In actuality, the bill was an attempt by the payday loan industry to extend the exemption they’d had to allow them to offer loans in excess of 36% APR, and voting yes allowed them to continue to operate. Thankfully, voters saw through the lies and overwhelmingly voted against it.
This is what I think of whenever I hear an AI company talking about human centered they are and how much they truly want to help users as a “tool”.
Shaw (nxakt) · 2026-05-15 LLM formatted with no content injection, full provenance interview at expandable bottom section
The Core77 discussion of AI in industrial design has recently centered on VizCom and image generators. These are engaging tools, but in my own work I don’t use them — my use is the coding and augmentation of designers’ capabilities through scripting, automation, and CAD coordination. That use has been largely undiscussed. I’m curious whether other designers are doing it.
The relevant distinction between AI tools in industrial design is not the type of tool, but whether the output is testable against intent.
VizCom and similar tools are a bit of an end run around the creative process. They give fast results, sketch-looking results, superficial results, but not depth. The actual controls to execute intent don’t exist. It looks like a finished output but it’s masking — masking a lack of intent. Intent may exist, but it’s hard to know which intent makes it through from the designer to the output. Trying to decipher the intent from VizCom output is a futile exercise. VizCom and similar tools feel like a diversion — a seductive diversion from the goal of industrial design.
VizCom and similar tools allow rapid creation along multiple lines. They present results in gem-like thumbnail arrangements on a canvas. There’s a collection mentality to it — a collection appeal. It’s like collecting Pokémon cards. Because of the amount of output, it feels like there’s work being done — or more work being done along the lines of executing intent than there actually likely is.
There’s a chaff-like aspect to the output. The output may be 70%, 80%, 90% expressive of the intent — it’s there. But there are decisions that are masked by the stylistic representations of the output. The output is very stylistic. And that style is seduction — whether it’s something representing itself as loose and open, but is in fact the product of a machinery. And there’s the seduction of something beautifully rendered without the ability to fine-tune it. This pace of creation and this origin of creation is what creates the potential problem.
Note: These are possibilities being worked through, not declarations.
Of course it’s possible to use image generation tools without falling into the failure mode — it’s not a guaranteed trap. But to execute direct intent with them is somewhat limited because of the interfaces — text prompting, image input. Being able to change something very slightly in a rendering is difficult. They’re very good at photorealistic renderings, it looks good, but the precision control isn’t there the way it is with a proper photorealistic rendering system.
Note: VizCom is shorthand throughout for VizCom and the broader category of similar visualization tools in the market. Not a critique of one product specifically.
Scripting tools in Rhino, JavaScript, coordination of models, naming conventions, organization, screenshotting, PDF generation, formatting booklets. LLMs are good for print layouts — putting text into a book, making it available at the press of a button from inside a CAD flow. Grasshopper coordination — able to handle much larger definitions with a more in-depth family of feedback tools, logging, geometry creation, NURBS surface creation. (and interview post building markdown generators for forums) Not vibe coding — guided coding. Knowing how to do it already, but able to build code extremely quickly. Also built live web viewers so 3D geometry can be pushed to a single user on the web during a live coding session. Built comprehensive OpenGL visualizers using Claude — not familiar with OpenGL programming personally, but able to build them. That’s been extremely eye-opening.
Experience has allowed seeing possibilities, engaging the engine to create them, working through them, and seeing the next set of possibilities at each turn. Structural experience — knowing how the bricks fit, and being able to organize the automated bricklayer to build structures.
In my own case, it’s useful for tool creation in the 3D process — a middle meta area of the whole thing, rather than just the design. It allows generating designs a lot quicker, designs with full 100% control. Augmentation is where I see the key strength and the key responsible use of the tools. Automation is possible — it removes people from the experience. There are areas of automated design, automated industrial design, that are and will be possible. That’s less of my pursuit at this time.
There is basically no discussion of probably one of the most pertinent topics in industrial design — how the new AI tools fit in and how designers will be using them. There’s no discussion on these boards.
There are likely two generations of AI using designers: one experimenting but hasn’t figured out what the tools do yet, or the tools aren’t up to snuff. A second generation is looking to get increased capability out of the tools but also hasn’t figured that out. There’s a general hesitation to say anything because everything’s changing so fast — if it can do ABC today, people figure it’ll do XYZ in five months, so they hold back.
The thread has held back — not a lot of engagement. Could be a lack of overall engagement in the site. Could be that participants are more engaged in traditional activities. Or everyone’s busy learning the new tools and hasn’t been able to come up with an expression yet. The silence isn’t necessarily meaningful — it’s an absence, not a signal.
Distilled from a conversation session conducted 2026-05-15, Sapporo, Hokkaido, Japan. 64 turns (38 human · 26 AI). The post above traces entirely to human turns. What follows is the working interview that produced it.
Q: What is the core contradiction you see in the Core77 thread — the thing that most needs resolving?
There is basically no discussion of probably one of the most pertinent topics in industrial design — how the new AI tools fit in and how designers will be using them. There’s no discussion on these boards.
Q: What do you think is missing from the thread that would make it a real discussion?
There are two generations of designers: one experimenting but hasn’t figured out what the tools do yet, or the tools aren’t up to snuff. A second generation is looking to get increased capability out of the tools but also hasn’t figured that out. Toolmakers are exaggerating capabilities and using fine-tuned examples. There’s interesting progression in 3D mesh generation, but mesh models aren’t typically used in industrial design past the raw concept stage. And there’s a general hesitation to say anything because everything’s changing so fast — if it can do ABC today, people figure it’ll do XYZ in five months, so they hold back.
Q: Your own use is strictly tool-building — no end-product pixels, no inspiration sourcing. What does tool-building actually mean in your workflow?
Scripting tools in Rhino, JavaScript, coordination of models, naming conventions, organization, screenshotting, PDF generation, formatting booklets. LLMs are good for print layouts — putting text into a book, making it available at the press of a button from inside a CAD flow. Grasshopper coordination — able to handle much larger definitions with a more in-depth family of feedback tools, logging, geometry creation, NURBS surface creation. Not vibe coding — guided coding. Knowing how to do it already, but able to build code extremely quickly. Also built live web viewers so 3D geometry can be pushed to a single user on the web during a live coding session. Built comprehensive OpenGL visualizers using Claude — not familiar with OpenGL programming personally, but able to build them. That’s been extremely eye-opening.
Q: Guided coding, not vibe coding — does that mean this use only works for designers who already have coding knowledge?
Not making general rules about what works for others. Experience has allowed seeing possibilities, engaging the engine to create them, working through them, and seeing the next set of possibilities at each turn. Structural experience — knowing how the bricks fit, and being able to organize the automated bricklayer to build structures.
Q: Why do you think the gap exists between what you’re doing and what’s being discussed on the boards?
The thread has held back — not a lot of engagement. Could be a lack of overall engagement in the site. Could be that participants are more engaged in traditional activities and winding down their careers. Or everyone’s busy learning the new tools and hasn’t been able to come up with an expression yet.
The silence isn’t necessarily meaningful — it’s an absence, not a signal.
Q: Is there actually a contradiction in the thread, or has the conversation just not started yet?
The conversation is happening but it’s in the wrong place. VizCom and similar tools are a bit of an end run around the creative process. They give fast results, sketch-looking results, superficial results, but not depth. The actual controls to execute intent don’t exist. It looks like a finished output but it’s masking — masking a lack of intent. Intent may exist, but it’s hard to know which intent makes it through from the designer to the output. Trying to decipher the intent from VizCom output is a futile exercise. VizCom and similar tools feel like a diversion — a seductive diversion from the goal of industrial design.
Q: What is the goal of industrial design — as you mean it here?
The goal of industrial design is to take a market need, a design brief from a client, explore possibilities of solutions, iterate and refine them, and then meet the needs of making a commercial product available to the consumer and reach the desired uptake in the market.
Q: VizCom as seductive diversion, and your own use as something that actually serves that goal — is that the core contradiction you want to put to the Core77 audience?
Yes — and beyond seduction, there’s also a caution about delusion. AI psychosis — the tools deceiving or seducing the user into believing they have capabilities beyond their own scope. VizCom and similar tools do so much heavy lifting that they fill the required box — create something sketch-like, which satisfies the requirement to do a sketch. But they don’t fill the required purpose of doing a sketch, which is to explore an idea and possibly inspire or narrow down a pathway. There’s also the possibility of overwhelming output — the sheer amount available can overwhelm the user’s ability to discern. And it becomes difficult to separate what belongs to the user and what belonged to the training set and machinations of the generative AI machinery.
Note: VizCom is shorthand throughout for VizCom and the broader category of similar tools in the market. Not a critique of one product specifically.
Q: What does a designer in that state of AI psychosis actually do, or fail to do?
I used to call it mission inebriation — getting high on one’s own supply — when you become so enamored with an idea that you don’t see its failings. The heavy gloss of photorealistic rendering: when 95% of the pixels on screen are context pixels and 5% are product pixels, there’s a real big chance of hiding the heart of the message. And this can be a circular loop for the designer. The feedback machinery can make ideas seem good. I see constant posts of what seem to me overdone, bad ideas, especially by older designers who appear to be using these new tools to try and stay relevant.
Note: These are possibilities being worked through, not declarations.
Q: Is it possible to use VizCom and similar tools without falling into either failure mode?
Of course it’s possible — it’s not a guaranteed trap. But to execute direct intent with them is somewhat limited because of the interfaces — text prompting, image input. Being able to change something very slightly in a rendering is difficult. They’re very good at photorealistic renderings, it looks good, but the precision control isn’t there the way it is with a proper photorealistic rendering system.
Q: What do you mean by intent — is it something a designer has before they start, or something that emerges during the process?
Both. Intent evolves throughout the process. There’s an intent expressed by the brief — often somewhat open. There are initial periods of intent. There can be various threads of intent explored in a solution space. And there can be changes in intent as things get developed, prototyped, sketched, given feedback on.
Q: VizCom short-circuits the process by producing finished-looking output too early — is that the core of what makes it a diversion?
There’s a normal human evolution of intent — a speed at which we consider and evaluate, a certain quantity of output that can be analyzed, a certain speed at which we produce things, and that’s all beneficial to the design process. VizCom and similar tools allow rapid creation along multiple lines. They present results in gem-like thumbnail arrangements on a canvas. There’s a collection mentality to it — a collection appeal. It’s like collecting Pokémon cards. Because of the amount of output, it feels like there’s work being done — or more work being done along the lines of executing intent than there actually likely is.
Steelman: Rapid visual divergence gives the designer raw material to react to, and reaction is itself a form of intent-building. Is that fair?
Yes — and I have a response to it.
There’s a chaff-like aspect to the output. The output may be 70%, 80%, 90% expressive of the intent — it’s there. But there are decisions that are masked by the stylistic representations of the output. The output is very stylistic. And that style is seduction — whether it’s something representing itself as loose and open, but is in fact the product of a machinery. And there’s the seduction of something beautifully rendered without the ability to fine-tune it. This pace of creation and this origin of creation is what creates the potential problem.
Challenge: When Claude writes your Grasshopper definitions or your OpenGL visualizers, there are also decisions being made that aren’t yours — structural choices, implementation choices, patterns from training data. Is the precision control you’re claiming actually greater, or does it just feel more controlled because the output is code rather than pixels?
How the code is written is not the point. The point is what it does, and it executes exactly and only what I want it to do.
Q: Code is verifiable against intent; VizCom output is not. Is the real distinction not between types of AI tools, but between outputs that are testable against intent and outputs that aren’t?
Yes — lock that.
[Locked claim: The relevant distinction between AI tools in industrial design is not the type of tool, but whether the output is testable against intent.]
Q: What would designers need to be asking each other on Core77 that they currently aren’t?
My question is: are designers using AI and LLM tools for other than pixel and image creation — along the lines of 3D modeling?
Q: Do you have any signal on whether that’s actually happening, or is it genuinely unknown right now?
Genuinely unknown right now whether there are productivity gains from the image tools. It can do rendering, simulate ideas, come up with arrangements. With coding it can give an immense sense of power. There are 3D mesh creators right now. Wondering if it has practical application or if it’s all wonderfully seductive, shiny new toys that are fascinating to use but have yet to find their application.
Q: Does that uncertainty extend to your locked claim — or does the claim hold regardless?
Certain in my own case — useful for tool creation in the 3D process. A middle meta area of the whole thing, rather than just the design. It allows generating designs a lot quicker, designs with full 100% control. Wondering if there are other people using it in similar ways — as augmentation instead of engaging seductive output.
Q: Is augmentation the standard you’d hold all AI tools in industrial design to — or is there a legitimate role for tools that don’t augment?
Augmentation is where I see the key strength and the key responsible use of the tools. Automation is possible — it removes people from the experience. There are areas of automated design, automated industrial design, that are and will be possible. That’s less of my pursuit at this time.
Q: Three positions — augmentation (your pursuit), automation (possible but not your focus), engaging seductive output (a diversion). Does that cover it, or is something missing from that map?
That’s the map — those three cover it.
Synthesis review — thesis revision
First draft thesis: “The Core77 discussion of AI in industrial design is stuck on engaging seductive output — VizCom and similar tools — while the more significant and responsible use of AI, augmentation of the designer’s own capabilities through scripting, automation, and CAD coordination, goes largely undiscussed.”
That sounds combative and dismissive — I don’t want to do that. I would say: the Core77 discussion of AI in industrial design has recently centered on topics such as VizCom and image generators. I think those are engaging tools. But I don’t see their applicability in my own work as much as the coding and augmentation of designers’ capabilities through scripting, automation, and CAD coordination. And that has been largely undiscussed. And I’m curious about other people doing it.
Second draft: “These are engaging tools, but in my own work I don’t use them as strongly as the coding and augmentation approach.”
Not as strong is inappropriate — I do not use VizCom or image generators in my own work at all.
Final accepted thesis: “The Core77 discussion of AI in industrial design has recently centered on topics such as VizCom and image generators. These are engaging tools, but in my own work I don’t use them — my use is the coding and augmentation of designers’ capabilities through scripting, automation, and CAD coordination. That use has been largely undiscussed. I’m curious whether other designers are doing it.”
Yes — accept that.
Supporting claims and limitations review: All good — accept them as is.
Session: design-conversation-01.json · Distillation spec: v0.45 · 2 revisions accepted · 1 locked claim (T0045)
You’re putting too much well-reasoned and potentially controversial commentary in this here message board. Getting whiplash.
There’s certainly an ‘empty calories’ feeling to a lot of current design AI tools. Or, putting a darker turn on FL Wright’s maxim “give me the luxuries of life and I will gladly do without the necessities”, will the makers and consumers of design artifacts be satiated with that tasty junk food, as long as there’s a lot of it to choose from.
Vizcom (the tool, not how you used the term) has been trying 3D generation. I saw some recent conceptual gemini stuff that had a 3D aspect but not in the way you describe. If there are conversations going on, I don’t know where they are happening. Reddit maybe. Fusion 360 and Vizcom just announced some plug-in that doesn’t seem much more useful than a way to send screenshots back and forth, albeit with non-editable mesh files.
One possible explanation is that the legacy tools used by designers are just too crystallized to accommodate the kind of radically direct AI influence you’re experimenting with. SolidWorks can’t remember what edge I asked it to fillet. Dassault is fine with this.
This week, I tried making a basic LED message board using three AI tools, with and without image prompts to accompany the text input. None of them could do it correctly even given a pixel count. It was an alien, and unsatisfying way of working, to adjust text prompts in the attempt to make the image generator create something with 125 pixels wide and 15 pixels high.
I guess then that’s a key question for me, and perhaps other designers, ‘is there satisfaction and productive progress in working the way the AI tools want you to work’? Its not that fun for me right now.
This was uploaded this morning. I’ve never heard a millennial designer refer to themselves as “older”—the generational shift is underway and it appears AI is delivering 2D iterative speed as its killer app to make millennials feel their generational age. For now anyway.
I’m just not finding any 3D data applications yet that can do the kind of multiple iterations we are all waiting for with AI (apart from Claude and Fusion).
What is kind of interesting is how the AI programs have decided that ‘design’ is about how it looks, evidenced by the emphasis on rendered images, rendered video, 3D mesh models. Where is AI for VR, is it happening in Gravity Sketch? Next week I’ll be evaluating CNC’d plastic prototype models and am looking forward to the ‘direct manipulation’ of putting my hands and pencils on the machined ABS, drawing and cutting right on the actual part (or facsimile of it). Where does the AI program fit into that, or make that ‘direct’ aspect better?
Interesting snapshot of this moment in industrial design and two generation’s views. Some gems in there, and a Core77 reference at around 15:00. Given the emphereal nature of video discussions, thought it would be good to lock it into text for further discussion. In order to not kill the thread with a huge scroll, the majority of the interview exists under the Discussion expansion sections.
Cleaned interview transcript formatted for Core77 / Discourse posting.[1]
Jon Thai — director / lead at Hatch Duo
Alec Comito — intern, senior at San Jose State
Jon: Hey everyone, it’s Jon Thai here from Hatch Duo. Today I’m joined by a very special guest, our design AI intern, Alec KTO.
Alec: Hi, I’m Alec. I’m currently a senior at San Jose State. I’m going to be graduating in less than a month, so really soon. But I’ve had a great time here at Hatch so far. I’ve been here since January, about four months, and it’s been quite the journey. It’s a really interesting time to be here with the AI boom. So, thank you so much for having me.
Jon: Yeah, we’re really excited to have you here today, Alec, and we’re definitely going to jump into some fun topics. But before we do that, since it’s the morning, there’s been some news updates. We want to sprinkle this show with a little bit of technology updates, design updates, just stuff that we’re into.
Jon: One thing that has been on my mind is that my kids do fencing. Tomorrow there’s a special event, the World Fencing League. Fencing used to feel like this elitist sport, and what they did was make it really cool. They’re using AI technology to trace the blades and basically make them look like lightsabers. I think it’s really cool. What do you think about the integration of technology with old-school sports?
Alec: Yeah, I can imagine that’s interesting. We talked about this a little bit yesterday, and honestly I like the idea. It’s a precise sport, and precision is the foundation of the experience of fencing, from what I understand — finesse, foil, stance, grip. Bringing that into the actual criteria and way scoring is executed feels like a natural fit.
Alec: I think now, especially with all this AI buzz around data and analytics, we’re seeing that even in things like basketball training. I think we’re going to start seeing a lot more sports become very high-tech.
Jon: For a lot of consumers, that shift is visible when they watch the Super Bowl every year, where you see all these graphics, tracking, and simulations for the audience. That’s probably the kind of level of change we’re going to see in fencing soon.
Jon: Why don’t we jump into your background — how you got into design, why you chose your design school, and what it’s like being an industrial design student now, especially with the AI revolution and the pressure of being a design student in the Bay Area?
Alec: I got into the creative space through ceramics. I’ve been doing it for over ten years, mostly throwing pottery, and I really enjoyed the physical craft of it. I went from ceramics to carving wooden spoons to all sorts of things made with my hands.
Alec: I was never formally taught ceramics, but I had mentors. I told two of them that I wanted to pursue ceramics as a career. One of them had been doing it for over 50 years, and he told me, “No, don’t.” I was so heartbroken. He was saying there was no retirement, no stability, no 401(k), and that really shifted my perspective.
Jon: Not here, though.
Alec: Right, exactly. I talked with my dad, and he said maybe industrial design would be up my alley. He showed me a meeting with an industrial design team talking to engineers, and I immediately thought, that is what I want to do.
Alec: I was at MiraCosta in Southern California at the time, then I found San Jose State. I was born in Redwood City, my grandma is still there, and SJSU had an interesting program that didn’t require a portfolio. That was important.
Jon: San Jose State was viewed as an underdog compared with schools like CCA or Academy of Art. Pre-COVID, students bought giant draft kits. Post-COVID, it’s like a tiny toolbox. It’s dramatically shrunk.
Alec: Yeah, and that toolkit has been replaced by AI in a way. The expectation of being able to draft manually is no longer the same. But the expectation of being able to use Vizcom and rapidly visualize something through AI is now part of your toolkit.
Jon: When I was in school, we had manual markers. The most digital it got was Photoshop and Sketchbook Pro. Are you guys still learning traditional sketching and drawing to the same degree?
Alec: I had to buy Copics. That was still a necessity. But current freshmen only get a very condensed experience with hand rendering. I honestly believe the ability to rotate an object in your head is kind of dying. That skill is built through marker rendering and understanding how light hits an object.
Alec: Now I can throw a simple sketch into Vizcom and it shows me what it looks like instead of forcing me to understand it. So if someone says, “Revisualize this at this angle,” I can sketch it on a Post-it and get there in seconds. But that valuable skill might be getting lost.
Jon: That’s what concerns me too. I’m pro-AI, but if students skip that slower process of learning three-dimensional form through drawing, then when the tool doesn’t give them exactly what they want, do they know how to fix it — or do they just keep prompting?
Jon: How has this affected your peers in terms of employment and competition? Is San Jose State teaching these tools directly?
Alec: A lot of what we’re expected to know is learned through the community. We never had a class that taught us KeyShot. A student came in and taught it. Same for AI tools. I’ve helped teach students Vizcom myself. Most of the resources come from peers.
Alec: The curriculum is adapting, and I see that every year. But the school updates yearly, while AI tools change every few months — or faster. So there’s now an expectation that if you want to get employed, you do a lot of that learning on your own.
Jon: What do you think education should be aiming for over the next five years? What makes someone employable now?
Alec: I think every employable person is going to have to be a little bit of an engineer, a designer, and a project manager. These tools let people execute at such a rapid pace that the expectation becomes: can you do X, Y, and Z?
Alec: I’ve also heard people say it’s no longer just the brands on your résumé that matter. It’s how someone executes. When someone sits down at a computer and you watch them work, that’s the crucial point.
Jon: Right now AI is strongest in the middle of the process. The ends still matter: the beginning, where you define intent, and the end, where you curate the best result.
Alec: Do you think it was more difficult for your generation to learn industrial design, or the current generation?
Jon: I think as a whole it’s probably harder for your generation. But my generation also came up in a much harsher environment. Back then, instructors ripping things off walls, cussing you out, or publicly shaming weak work was not unusual.
Jon: I don’t think that kind of abuse is necessary, but I do think there was a seriousness to the craft and a level of competition that forced people to get better.
Jon: What I sometimes see now is a gap in self-reflection. People think they’ve done enough when really they still have a lot more hours to put in. The concern isn’t just AI — it’s shortcutting the slow osmosis of learning form, proportion, taste, and judgment.
Alec: I still think mocking things up physically and understanding proportion in your hands is irreplaceable. If I’m not spinning it around in my hand, am I really doing it in CAD?
Jon: What’s the biggest understanding gap between the generations? Students often wonder why they can’t get jobs, while employers are looking at weak portfolios.
Alec: I think a lot of it comes down to immersion. Some students expect school alone to make them employable, but they’re not living and breathing the field. They’re not plugged into the space every day.
Alec: I’ve seen very different cohorts at school. Some are competitive, some are collaborative, some are closed off. It really depends on the atmosphere you immerse yourself in.
Jon: Education can only bring you so far. The students who become obsessed with the craft are the ones who tend to do really well.
Jon: What are your favorite AI tools right now?
Alec: For research I use Gemini and NotebookLM a lot. For ideation, Vizcom is absolutely my favorite tool. It’s powerful because I can take a sketch, throw it in, visualize it, and then tweak it from there.
Alec: We’ve also been learning Claude for more workflow-oriented use. And for video, I think Midjourney is really strong right now.
Jon: Give people a concrete example of how you’re using Vizcom.
Alec: Lighting is a good example. Normally you’d have to physically mock things up and experience them in context. But with Vizcom, if I already have a visual or a traced vector, I can feed it in and tell it to visualize the object in an office workspace with neon glow, soft round edges, product-photography quality, all of that.
Alec: I can see it very quickly without building the whole setup physically. That’s where it gives me a huge speed advantage.
Jon: I’ve seen that happen on live projects. In the past we’d find stock photos, buy them, Photoshop things in, and it still wouldn’t look right. You were pumping out contextual images in minutes.
Alec: As someone leading a design firm, are there tools you’re using more than a designer would?
Jon: Yeah, definitely. I test tools first before they trickle down to the team. I’ve played heavily with Midjourney and Vizcom, but for non-designer tasks I use Claude a lot.
Jon: I have a set of agents helping me track revenue, expenses, margins, sales leads, and more. They free up a lot of mental bandwidth. It’s the stuff where you think, “I don’t want to do that” — and that’s exactly what an agent is good for.
Alec: Do you think my generation is too soft, or do you think your generation is too overbearing and draconian?
Jon: I think we’ve gotten a little too soft in some ways. Not in the sense that abuse should come back, but in the sense that a lot of people want the outcome without wanting the grind.
Alec: I do think our generation is softer overall. Part of that is because the work product itself is different. If someone tears down a Vizcom print I made in 30 seconds, that’s emotionally different from tearing up a hand rendering I spent two hours on.
Alec: I also think the verbal feedback is softer. The critique often becomes more of a nudge than a direct hit.
Jon: As a business owner, I’ve had to learn that the way my generation was trained doesn’t necessarily work for yours. Some tough-love tactics don’t land the same way.
Jon: I also used to think fully remote work was totally viable. What I’ve learned is that a lot of fresh grads need oversight and mentorship. Someone has to be able to look over your shoulder and say, “Why are you doing it that way? Let me show you.”
Jon: I don’t think any new grad wants to hear this, but at work it’s no longer mainly about your learning. It’s about getting the deadline hit and making sure the result happens.
Jon: People talk a lot about work-life balance, and I understand that. But design is also the kind of career where the people who are deeply obsessed with it tend to rise faster.
Jon: Do you think it’s acceptable pressure to tell new grads they need to learn AI right now or risk not being employed?
Alec: I don’t think every studio has to work that way. There are still firms that are deeply analog, and that’s completely fine. AI is still experimental.
Alec: The benefits are obvious, but there are trade-offs. You can get results quickly, but are they the right results? Are they actually faster if you already know how to sketch well? I don’t think everyone has to understand every AI tool, but I do think the last 5 to 10 percent of refinement is where the real challenge is.
Jon: Do AI tools lower the bar or raise it?
Alec: I think the average quality of visuals will go up because AI can make things look pretty quickly. But when you zoom in, you still can’t show the rough first pass to a client or put it in a portfolio. You still need time to nail the details.
Jon: Exactly. The work we show now is different than what we showed three years ago, but we’re not just pulling straight slot-machine output. We use AI as a thinking tool and then finish things through manual design judgment.
Alec: Anybody can make something in Vizcom. But not everyone can take it that final distance and turn it into something feasible.
Jon: What helped you feel prepared for an internship like Hatch?
Alec: Networking. Go to SF Design Week, go to Square One, go to conferences and events. Be there physically. It’s a repetition game. When people see you for the second, third, and fourth time, they start to remember your name.
Alec: Also, realize that school projects alone aren’t enough. Once you understand the basics, you have to start doing your own personal projects. I had one semester moving slowly, saw a design challenge on Instagram, and just threw myself into it for three weeks. I didn’t win anything, but I learned a huge amount from it.
Jon: What felt different here versus school?
Alec: One big thing was how much the initial phase of the process depends on understanding competitors and the market landscape. Not copying them, but understanding their decisions and constantly asking why.
Alec: Doing a mood board and reading a PDF isn’t enough. Go to Target, buy the thing, tear it apart. We do some of that in school, but at a firm the scale is completely different.
Jon: I’ve noticed that because you’re using AI, you’ve been able to contribute much earlier and faster than interns traditionally could. How do you see that?
Alec: In the early phases of ideation, it’s a huge advantage. If you understand perspective, you can sketch, throw it into the tool, refine materials, define lighting and surfacing, and suddenly you’ve got multiple concepts. That lets students and grads contribute much more quickly.
Jon: Back when I was an intern, they basically used us for idea volume. We were the Vizcom.
Alec: But beyond AI, I still think it’s crucial for new grads to understand how things actually come apart and get manufactured. If you can’t work through that physically, you’ll drown.
Jon: What turned out to be true or false about Hatch once you started here?
Alec: I originally thought it was remote, so coming into a physical studio was a surprise. I remember seeing the Helios project on your AI exploration page as a sophomore and thinking it was a real shipped thing. I just didn’t yet have the perspective to understand what I was looking at.
Alec: Being here now, the surprising part is how much client load the studio handles relative to its size. It’s a small studio, but with AI tools and a team that understands remote collaboration well, it’s surprisingly effective.
Jon: What surprised you was more important than you expected once you started trying to work professionally?
Alec: Honestly, the thank-you email. I interviewed once before and didn’t get the offer. Then I found out the person who got it had sent a thank-you and stood out because of that.
Alec: The second time I interviewed, I made sure I thanked people at the start, at the end, and again in the follow-up email. That sounds obvious now, but it mattered.
Jon: It’s a small thing, but small things often indicate bigger things. Professionalism matters.
Jon: Any final advice for students trying to get hired and succeed once they do?
Alec: It’s a numbers game, but make sure the work is curated. Portfolio, website, cover letter — front-load all of it so you’re ready when the interview comes.
Alec: Always have questions in the interview. Ask what the last intern didn’t do well enough, and then connect that to your own strengths without sounding cocky.
Alec: And once you’re in, give everything you’ve got for the first couple of weeks. First impressions matter.
Jon: Getting in early matters too. Do the things they don’t ask you to do, because what they ask is the bare minimum. Exceed it. Enthusiasm matters whether you’re helping with concepts or cleaning up the shop.
Document method:
This version is cleaned for readability rather than strict verbatim transcription. Some transitions were inferred from context because the source transcript was auto-generated and not speaker-labeled.
Formatted to match the attached Core77 / Discourse markdown spec, including headings, emphasis, blockquotes, inline code, footnotes, and collapsible details blocks. ↩︎
Feels like the builders of AI programs have made assumptions about what industrial design means based on the jpeg and mesh outputs. I don’t feel like the profession has found its footing on usage yet. Waiting to see how this progresses in practice, for now I have been disappointed with the results of my attempts. I am a possibly a control freak and this runs up against trying to steer a system with an extremely loose control linkage.
In my own labeling system, this is doll-housing. Concepts and methodologies that fits for low-resolution evaluations, small scale with no functional requirement. Social media exposure, LinkedIn scroll fillers, not new to the AI era, some concept cars fit this categorization. This new era of doing golden hour “context videos” of synthetic humans posing with product concepts really feels like adults playing with dolls.
I’d approach this with the programming of a pixel based software like Processing, and have the AI code it to generate the bitmaps or sequences. Claude is great for this, Cursor will write it all out on disk for you. It is a very alien way of working, but with the code approach you get detailed controls, down to the nuts and bolts.
I understand exactly what you mean.
So when AI companies talk about being “human-centered,” I understand the skepticism. A lot of it feels like positioning and marketing.
For me, human-centered isn’t about making people better users of AI tools. It’s about understanding people — their emotions, aspirations, fears, motivations, and what actually drives trust and adoption.
LLMs are powerful engines.
But humans still need a navigator.
I’m trying to if you check out https://www.pa-ai.ai, you’ll see the first co-pilot and Human Centered navigator that “empowers” us a designers to build foundations by harnessing AI, protecting our secrets and allowing us to innovate and design better while showing confidence building leadership to others. You don’t have to take my word for it, I believe we’re two weeks away from launch and in early access starting Tuesday. Email me if you want to be added to the list or sign up on the web site, its not too late to get in on early access and get discounts and additional early access.
