What AI is Really For
Best case: we’re in a bubble. Worst case: the people profiting most know exactly what they’re doing.
After three years of immersion in AI, I have come to a relatively simple conclusion: it’s a useful technology that is very likely overhyped to the point of catastrophe.
The best case scenario is that AI is just not as valuable as those who invest in it, make it, and sell it believe. This is a classic bubble scenario. We’ll all take a hit when the air is let out, and given the historic concentration of the market compared to previous bubbles, the hit will really hurt. The worst case scenario is that the people with the most money at stake in AI know it’s not what they say it is. If this is true, we get the bubble and fraud with compound motives. I have an idea about one of them that I’ll get to toward the end of this essay. But first, let’s start with the hype.
As a designer, I’ve found the promise of AI to be seriously overblown. In fact, most of the AI use cases in design tend to feel like straw men to me. I’ve often found myself watching a video about using AI “end to end” in design only to conclude that the process would never work in real work. This is usually because the process depicted assumes total control from end to end — the way it might work when creating, say, a demonstration project for a portfolio, or inventing a brand from scratch with only yourself as a decision-maker. But inserting generative AI in the midst of existing design systems rarely benefits anyone.
It can take enormous amounts of time to replicate existing imagery with prompt engineering, only to have your tool of choice hiccup every now and again or just not get some specific aspect of what a person had created previously. I can think of many examples from my own team’s client work: difficult to replicate custom illustrative styles, impossible to replicate text and image layering, direct connections between images and texts that even the most explicit prompts don’t make. A similar problem happens with layout. Generative AI can help with ideating layout, but fails to deliver efficiently within existing design systems. Yes, there are plenty of AI tools that will generate a layout and offer one-click transport to Figma, where you nearly always have to rebuild it to integrate it properly with whatever was there beforehand. When it comes to layout and UI, every designer I know who is competent will produce a better page or screen faster doing it themselves than involving any AI tool. No caveats.
My experience with AI in the design context tends to reflect what I think is generally true about AI in the workplace: the smaller the use case, the larger the gain. The larger the use case, the larger the expense. Most of the larger use cases that I have observed — where AI is leveraged to automate entire workflows, or capture end to end operational data, or replace an entire function — the outlay of work is equal to or greater than the savings. The time we think we’ll save by using AI tends to be spent on doing something else with AI.
(Before I continue, know also that I am a co-founder of a completely AI-dependent venture, Magnolia. Beyond the design-specific use cases I’ve described, I know what it means to build software that uses AI in a far more complex manner. The investment is enormous, and the maintenance — the effort required to maintain a level of quality and accuracy of output that can compete with general purpose AI tools like ChatGPT or even AI research tools like Perplexity — is even more so. This directly supports my argument because the only reason to even create such a venture is to capitalize on the promise of AI and the normalization of “knowledge work” around it. That may be too steep a hill to climb.)
Much has already been made of the MIT study noting the preponderance of AI initiative failures in corporate environments. Those that expect a uniform application of AI and a uniform, generalized ROI see failure, while those who identify isolated applications with specific targets experience success. The former tends to be a reaction to hype, the latter an outworking of real understanding. There are dozens of small-scale applications that have large-scale effects, most of which I’d categorize as information synthesis — search, summarization, analysis. Magnolia (and any other new, AI-focused venture) fits right in there. But the sweeping, work-wide transformation? That’s the part that doesn’t hold up.
Of course, we should expect AI to increase its usefulness over time as adoption calibrates — this is the pattern with any new technology. But calibration doesn’t mean indefinite growth, and this is where the financial picture becomes troubling. The top seven companies by market value all have mutually dependent investments in AI and one another. The more money that gets injected into this combined venture, the more everyone expects to extract. But there has yet to be a viable model to monetize AI that gets anywhere close to the desired market capitalization. This is Ed Zitron’s whole thing.
... continue reading