Insanely good. You already know this but I nurture the same fears in my darkest hours (so, all the time). The machines are coming to fuck us up, and it will be cosmic justice for thinking they could replace blue collar auto workers but never our white collar selves. We’re going to get rekt.
Brimming with thoughts and questions but of course feel no obligation to respond to any!
1. It’s really funny you still envision a front-end engineer! Why wouldn’t the Clippy of the future also spit out a functional front end, assembled from off-the-shelf SwiftUI bits? I think they’re kaput too.
2. This “omnipotent executive” is really a tragic figure if you play it out. I imagine a lone “ideas guy,” ordering his laptop to draw products, build them, write the marketing, and monitor the results. Is this really what we believe is the future of company building? Where is the teamwork? Where is the camaraderie? Where is the emotional support and varied expertise and human texture? Similar to arguments about embodied cognition - which reason that intelligence can never emerge in a vacuum, without a physical body making contact with reality - I am skeptical that an “executive” so insulated from humanity could ever accomplish anything great.
3. But let’s assume that’s how it is. Companies are now no-code, extremely lean, extremely centralized. You can make and deploy any app with the right spoken prompt. Well, wouldn’t this dramatic bar-lowering just create MASSIVE amounts of competition? And in the face of such competition wouldn’t people try to differentiate themselves? Perhaps via (you guessed it) a uniquely human approach to systems, design, and craft? It seems this whole cycle could in fact increase the value of design, though it at first seems to obviate it.
4. Along those same lines - I wonder what new feedback loops will appear on the consumer side of all this. I wonder if ai-gen apps will just be seen as unbelievably cringe. The type of thing you might ship to get signal but never hang your brand on.
5. The biggest gap I see between current state-of-the-art LLMs and what we’d need to overtake design is real-time local (read: intra- and inter-company) knowledge. A lot of “making the right call” in design depends on knowing you your company’s internal capabilities, as well as the external state of the market, and then also having some fuzzy and ever-evolving model of your customer’s needs. AI currently can not do any of these very well at all, and I think it would take a huge step change to get there. It would have to be in every meeting, see every slack, attend every offsite, internalize lessons from the past, etc. They might get there - I can imagine one day each company having a “personal” GPT trained on internal data. But for now at least I feel secure knowing that while FigmaGPT might be able to mock up a checkout screen (which is dope as hell!!!) it definitely can not mock up a check out screen for substack that delicately balances conversion, retention, tone, choice architecture, writer sensitivities, branding, and valuable data collection. In fact, it has no internal model of what these things are at all! These bots are worse than the worst dribbbler, utterly glib pixel pushers, and in conclusion they can all eat shit. We humans are here to stay.
I find myself wondering how much longer we’ll be thinking about software primarily in terms of apps that are developed and made available by businesses anyway. As the tools to create and modify arbitrary software are ultra-democratized, will we end up with OSes comprising a starter set of basic apps and a soupy layer of AI assistance? This can make a good-enough version of any utility app you can imagine, esepecially because you can mutate it on a whim any time it does something differently from how you expected it. For networked use cases that currently depend on some social service, decentralized platforms may seem more and more appealing because you can more easily mutate the software that reads them to your needs: just ask “what is Mills up to these days?”, “show me my mega-feed of everything; no, get rid of these ads; I like the music videos you’ve been finding lately, more of that please”, or “can I get a button that just saves stuff to a list to read later? cool, put that on everything from now on.”
One idle dream I have is: what if AI gets good enough that it makes architecture viable again? If it gives architects enough leverage, maybe everyone could go back to having a bespoke house (in this fantasy the AI has also done away with the board of supervisors)
Thanks for writing this. It's a really interesting take. Some questions emerged for me while I was reading. I'm already up way past my bedtime so I'll just list them.
Why are designers subject to commodification but not engineers and PMs?
Are architects in less demand? Compared to when?
Are we approaching some hard or soft limit on new software projects? We could have a decline in designers required per software project and still be fine if we have an increase in new projects undertaken. There seems to be lots of argument that AI will cause this trend (many new software projects).
Are architecture and software design similar enough to be subject to this commodification force in similar ways?
What is ultimately meant by commodification here? Is it something like "the consolidation of formerly bespoke stuff into more efficient but slightly worse and relatively interchangeable blocks of stuff"? Why aren't former architects finding work on higher level bespoke stuff composed of commodified blocks?
while people are arguing in the bird app about the tension/overlap between a designer and PM's responsiblities, Mills paints a clear picture of the more serious looming threat... appreciate this perspective
Insanely good. You already know this but I nurture the same fears in my darkest hours (so, all the time). The machines are coming to fuck us up, and it will be cosmic justice for thinking they could replace blue collar auto workers but never our white collar selves. We’re going to get rekt.
Brimming with thoughts and questions but of course feel no obligation to respond to any!
1. It’s really funny you still envision a front-end engineer! Why wouldn’t the Clippy of the future also spit out a functional front end, assembled from off-the-shelf SwiftUI bits? I think they’re kaput too.
2. This “omnipotent executive” is really a tragic figure if you play it out. I imagine a lone “ideas guy,” ordering his laptop to draw products, build them, write the marketing, and monitor the results. Is this really what we believe is the future of company building? Where is the teamwork? Where is the camaraderie? Where is the emotional support and varied expertise and human texture? Similar to arguments about embodied cognition - which reason that intelligence can never emerge in a vacuum, without a physical body making contact with reality - I am skeptical that an “executive” so insulated from humanity could ever accomplish anything great.
3. But let’s assume that’s how it is. Companies are now no-code, extremely lean, extremely centralized. You can make and deploy any app with the right spoken prompt. Well, wouldn’t this dramatic bar-lowering just create MASSIVE amounts of competition? And in the face of such competition wouldn’t people try to differentiate themselves? Perhaps via (you guessed it) a uniquely human approach to systems, design, and craft? It seems this whole cycle could in fact increase the value of design, though it at first seems to obviate it.
4. Along those same lines - I wonder what new feedback loops will appear on the consumer side of all this. I wonder if ai-gen apps will just be seen as unbelievably cringe. The type of thing you might ship to get signal but never hang your brand on.
5. The biggest gap I see between current state-of-the-art LLMs and what we’d need to overtake design is real-time local (read: intra- and inter-company) knowledge. A lot of “making the right call” in design depends on knowing you your company’s internal capabilities, as well as the external state of the market, and then also having some fuzzy and ever-evolving model of your customer’s needs. AI currently can not do any of these very well at all, and I think it would take a huge step change to get there. It would have to be in every meeting, see every slack, attend every offsite, internalize lessons from the past, etc. They might get there - I can imagine one day each company having a “personal” GPT trained on internal data. But for now at least I feel secure knowing that while FigmaGPT might be able to mock up a checkout screen (which is dope as hell!!!) it definitely can not mock up a check out screen for substack that delicately balances conversion, retention, tone, choice architecture, writer sensitivities, branding, and valuable data collection. In fact, it has no internal model of what these things are at all! These bots are worse than the worst dribbbler, utterly glib pixel pushers, and in conclusion they can all eat shit. We humans are here to stay.
I find myself wondering how much longer we’ll be thinking about software primarily in terms of apps that are developed and made available by businesses anyway. As the tools to create and modify arbitrary software are ultra-democratized, will we end up with OSes comprising a starter set of basic apps and a soupy layer of AI assistance? This can make a good-enough version of any utility app you can imagine, esepecially because you can mutate it on a whim any time it does something differently from how you expected it. For networked use cases that currently depend on some social service, decentralized platforms may seem more and more appealing because you can more easily mutate the software that reads them to your needs: just ask “what is Mills up to these days?”, “show me my mega-feed of everything; no, get rid of these ads; I like the music videos you’ve been finding lately, more of that please”, or “can I get a button that just saves stuff to a list to read later? cool, put that on everything from now on.”
You know the architecture thing kills me.
One idle dream I have is: what if AI gets good enough that it makes architecture viable again? If it gives architects enough leverage, maybe everyone could go back to having a bespoke house (in this fantasy the AI has also done away with the board of supervisors)
Thanks for writing this. It's a really interesting take. Some questions emerged for me while I was reading. I'm already up way past my bedtime so I'll just list them.
Why are designers subject to commodification but not engineers and PMs?
Are architects in less demand? Compared to when?
Are we approaching some hard or soft limit on new software projects? We could have a decline in designers required per software project and still be fine if we have an increase in new projects undertaken. There seems to be lots of argument that AI will cause this trend (many new software projects).
Are architecture and software design similar enough to be subject to this commodification force in similar ways?
What is ultimately meant by commodification here? Is it something like "the consolidation of formerly bespoke stuff into more efficient but slightly worse and relatively interchangeable blocks of stuff"? Why aren't former architects finding work on higher level bespoke stuff composed of commodified blocks?
while people are arguing in the bird app about the tension/overlap between a designer and PM's responsiblities, Mills paints a clear picture of the more serious looming threat... appreciate this perspective