AI is here to stay and is redefining the landscape of personal productivity, and I think that it is also transforming the very nature of human endeavour and striving. I am not the the first to have noticed how the use of AI coding tools creates a dopamine hijack in software engineers or vibe coders. Neither am I the first to assert that we're drowning in AI generated slop, be in written form, videos, audio and podcasts, or anything else.
More interestingly, there is a pattern on the part of users of AI to utilize AI as a substitute for thinking on their own. I say this as someone who has done exactly this. We live in a time of radical delegation, not just of the physical labour that made our muscles and bones strong, but of the mental labour that made our minds strong and that made us the species we are.
I don't know if we were promised greater productivity and human thriving by the tech boffins who built AI systems, especially the leading minds in the space who built and pushed ChatGPT, Claude and the like to the masses. It behooves us to examine their intentions more closely, and I say this as someone who is a practitioner of AI, and who has built models, and built on top of large language models, and the so-called foundation models. What is discernable is that these tools provide a specific kind of advantage to those who use them. Whether it is writing, coding, painting or other cognitively demanding tasks, the creation of content in its entire spectrum is covered by AI. Movies are being made with AI. It is possible for a person with just an imagination and the intent to use these tools to sequentially explore an idea for a film with AI, write a script with AI, prepare clips or scenes of video with AI, and edit and package them into a coherent film, also with AI. The user here provides the steering function, that guides the potential of the AI to some kind of impact.
To a technologist, this sounds like a promising application of AI. The practical purpose of embodying intelligence into formerly inanimate things around us, is to enable them to be animate in specific and helpful ways. Perhaps one path along the garden of forking paths in this general direction is the use of AI for cognitively demanding tasks. If AI could be used to solve hard mathematics problems, for example, some of us would find the weight lifted from our shoulders as we moved from trepidation and tedium to understanding. This is the logical positivist's spin on artificial intelligence, as a force for doing good to those that deserve that good. We know that good is subjective, though, and we will discuss this in this post.
To an AI technologist, the process of using gradient descent to recursively improve factor-response models on arbitrary data to learn a useful model ("useful" in the terminology of W. Edwards Deming) is a beginning. To make this system write passably good, human-like essays or have it generate incredibly creative and interesting images and videos is progress, and can be considered a technological or computational accomplishment. The fact that we have been able to sequentially transform a field of noise into a meaningful image, that is steered by a text prompt, to produce an image - this is no mean feat. And stacking such images together, nay, generating that sequence so that moving objects may be rendered in a form that we humans recognize as movies and clips - this is incredible too. In a nutshell, there is no doubting the technological achievements that these models represent. This may even represent a success - and I say "may even" for a reason.
I have described a beginning and progress in Henry Ford (beginning-progress-success) terms, so what then is the true definition of success in the use of artificial intelligence? To understand this, we need to understand the human perspective on artificial intelligence. Humans look to AI from the point of view of utility, ultimately. Even if the underlying techniques, methods, mathematics and statistics are explored from an information theoretic perspective and despite the fact that the theory and practice here is valuable, the underlying motivations have been the need to advance some agends of either humanity as a whole, or sub-sections of human society. There is a utilitarian version of the anthropic principle of sorts that I will have to invoke in this context. The anthropic principle the supposition that the universe created life so that it could observe itself and learn about itself - and I'll go out on a limb and say, measure its own utility. Similarly, perhaps biological intelligences create artificial intelligence so that they can learn about their utility. The bounds of artificial intelligences we create are not limited to the bounds of humans, however, neither is it the case that the AI we build are inside the closed systems we represent. Unlike the universe which subsumes life and artificial intelligence, humans share a landscape with artificial intelligence. Considering in this specific way the utility of artificial intelligence systems is a contrast to the libertarian, almost phenomenological view that a lot of business leaders and technologists have, in the context of artificial intelligence. Many of them anthropomorphize AI, and many attribute qualities to algorithmic systems that appear intelligence from a human standpoint. I impute an anti-phenomenological, utilitarian bias deliberately here, because phenomenology is not a good underpinning to reason about tools, specifically because tools are built with utility in mind.
Let's now think about what the human endeavour looks like, for someone writing, painting, or making movies. In the past, a writer would have to think through ideas from their experiences, from conversations, from books and articles they read, from shows they watched or songs they listened to. These were the "prompts", the inspiration for writing, for creating and for building ideas on top of old ideas and notions. They had to grapple with the unsafe silence their own minds would throw at them, when they came up with new ideas. They had to grapple with the uncertainty, of whether the idea would be valid or not. And they had to articulate these thoughts, however incipient or inchoate, so that (a) they can make sense of the ideas themselves (the reasoning utility), (b) they can explain it to others (the commuinication utilty). The reasoning utility for a writer of fiction may constitute the characters, their interactions, the plot and the denouement. These components helped them think about how a piece of fiction could be written. The communication utility in writing for a fiction writer would be their craft. Their use of the words and phrases, their descriptive toolkit for characters and interactions between them, the narrative style, the style of writing that drew people in. Some writers may be inventive, conceptualizing their own words, inventing their own languages, imagining the imperfections and quirks of their characters. The field of creativity is vast for human writers. Our brains are hallmarks of a general biological intelligence, and powered by language, reasoning, experiences that are multi-sensorial and the power of manifestation, we bring characters, ideas, places and the like, to life.
For the painter, the visual field of the canvas or paper, or other mediums is a blank slate. Of course, we all take inspiration for what we may want to draw and paint, but everyone approaches a scene or a subject differently. The painter's mind imagines a scene and the medley of paint and tools and brushes and the like translates the imagined scene into an approximation of the scene on the medium. The reasoning utility for the painter comprises of the subject of the painting, the composition and how subjects and objects are positioned on the visual field, and the broader meaning or import of the art. The communication utility of the painter are the strokes and the technique and the tools and the sequence of use of the tools and colours and the rest on the medium. In a similar vein, someone making a movie may find the reasoning utility of their movie to be the subject, the denouement and the plot with the characters and the like, similar to a book, while the communication utility here is the synthesis of actors, props, camera and screen play, effects, post-processing and editing. These together help tell the story that the creator intended.
Now let's examine the mental effort of these creators. They each have to begin from priors, and learn to nurture the ideas they work with, as they develop the reasoning around the ideas. The more they work and the more they produce, be it books or art or movies, the more they explore the landscape of tools and possibilities available to them, and the better they're able to respond in (a) confidence, (b) skillful execution, (c) speed and (d) cross-pollination of capabilities and ideas across mediums. What I discuss here is not ground-breaking, in any way, and is something anyone who has practiced any skill understands.
Coding and writing software are also similar. Tech professionals and knowledge workers are now getting used to using AI and prompting AI models to build applications, perform data analyses and deliver results. The human effort required to come up with a new logo, a new design for a report, a good opening paragraph, a good visualization that tells a story with data, a narrative that is compelling across a report, a piece of software that is built to spec but also performant, a beautifully made short movie for an ad in one or a couple of takes - these are likely to be replaced by AI that is prompted to produce different kinds of results. The possibilities with AI are remarkable. We have a tool that leverages the collective styles, ideas, notions and knowledge of billions of humans that came before it, and which can be prompted to give us what we desire, without the challenge and the pain of thinking about the ideas that much, and with no skill.
The radical delegation I mentioned in passing above is now able to be described in more exact terms. Radical Delegation is a pattern of behaviour where a human or a group of humans who's exceptionally skilled in a task decides voluntarily to delegate the process of executing that task to another human, group of humans, or a machine or a group of machines, that can perform the same task, in either the same fashion, or in an altogether different fashion. Radical delegation is not the same as automation but is a step on the path towards an industrialized or automated task. It is on the path to commoditization, either by the creation of human scale or mechanization of a task. Radical delegation is voluntary, and not imposed due to external constraints. When a software engineer decides to use AI to solve a problem, it is a voluntary choice for many at the moment. It may not stay that way for long, and there will be equivalent of a corporate program soon if not already present, in most teams that encourage the use of AI coding tools.
Beyond a point, the use of AI to automate work may not be an individualist question anymore, but one of either collectivism or interventionism. I say this because the questions that artisans or individuals ask about whether to use AI tools and how to use them will anticipate the questions asked by their managers, or people that deliver business or livelihood opportunities to them. At that point, the question becomes not one of choice, but one of efficiency, economics and markets. We've now moved from the field of free will (in the limited context of the use of AI tools based on one's arbitrary wishes), to the collectivist or interventionist approach to using AI, as a social policy which unfolds in the form of specific use cases, risks, and introduces both efficiencies and frictions in the form of the use of AI by humans in different contexts.
Collectivist culture in the context of AI on the internet takes many forms. We've seen movements like effective accelerationism (e/acc) become a force in prior years. This movement has advocated the use of technology in an interventionist fashion, primarily with the express objective of achieving greater human progress. The definition of what that progress entails is usually a little less well defined than the spirit in which technology and positivism is framed by e/acc. My own sense is that e/acc is akin to the culture that inspired tech companies in the 2000s and 2010s, a la "move fast and break things" and less akin to a serious movement which has well considered and rational use of technology to various ends. One could perhaps make a similar argument for AI tool adoption, in that there is a lot of depth in method, but less depth in motives. This implies a greater emphasis on communication utility, and a reduced focus on reasoning utility, for the concepts that e/acc engages with.
AI promoters and detractors for each of the various shades of automation exist. This spectrum of AI automation ranges from merely minor automations and thinking aids such as using tab-autocomplete features in code editors, on one end, all the way to full automation of the software engineering lifecycle on the other end. The promoters often state efficiency and the possility of moving grunt work away from humans and towards machines, as a noble end state. It is rare to see promoters acknowledge some of the pitfalls of this approach, but many do. For example, is the human a mere vessel with which to execute a software workflow? In the language of software engineering, are humans using AI mere wrappers with gates and checkpoints?
The detractors have both simplistic and more nuanced arguments. Detractors argue about skill atrophy, akin to how muscular or other atrophy is frowned upon by atheletes or artists. Detractors argue about how it is important for human expertise to survive and thrive, which is certainly noble. On the other hand, detractors don't often acknowledge the mundaneness and hardship associated with mastery and expertise. Here I refer in spirit to George Leonard describes the long plateau where "nothing happens" in his book Mastery, who acknowledges that the path to mastery of any skill includes a non-trivial amount of "grinding". This is the long plateau before we become adept at something, whether that is writing, painting, movie making or anything else. Detractors are acutely aware of the importance of reasoning utility, but not aware of the need for communication utility. I risk oversimplifying when I say this, and when I assert that detractors of AI automation don't acknowledge the need for communication utility arising out of broad adoption. Such broad adoption is precisely what drives progress in AI, which is known to be a data-intensive, compute intensive field that relies on ever more data to keep the train of AI model progress chugging along.
What does utopia look like, to the detractors, if the dystopia looks like radical delegation? Perhaps a term that could describe the opposite of radical delegation to AI, is personal thriving. By this I mean a condition where the human being continues to grow in judgment, in skill, in confidence, in taste and in responsibility, even while using tools that increase their leverage. Personal thriving is not Luddism. It is not a refusal of tools. It is not a romantic attachment to doing everything by hand, merely because it was done that way in the past. It is instead the insistence that the human being remain the site where understanding accumulates, where responsibility resides, and where the final meaning of work is interpreted.
This distinction matters because many discussions about AI collapse into a binary. One either embraces automation and scale, or one rejects them and becomes a reactionary purist. Reality is not so simple. A writer may use a thesaurus and still write. A programmer may use a debugger and still understand the code. A musician may use software instruments and still compose. In the same spirit, someone may use AI and still preserve their own thought. The question is not whether tools are involved. The question is whether the tool merely extends the person, or begins to displace the person's own reasoning.
The most dangerous form of displacement is not immediately visible in the produced output. This is what makes the question difficult. If AI produces a decent essay, a useful piece of code, a visual that is aesthetically appealing, or a summary that appears cogent, then the user is tempted to say that the process has succeeded. But this is only the communication utility being satisfied. Something legible has been emitted into the world. Something fit for consumption now exists. What has happened to the reasoning utility of the user in the process, however, is a separate question altogether. Has the user deepened their understanding of the problem? Have they become more capable of independently solving a related problem in future? Have they developed taste, judgment and character? Or have they merely learned to manage a prompt loop?
This, in my view, is the accountability problem.
When a human being performs a difficult task, they are not merely delivering an output. They are also shaping themselves. The action of struggling through a proof, of debugging a stubborn failure, of finding the right paragraph to close an essay, of understanding a difficult passage in a book, of revising an argument that does not yet cohere - these are not only means to an end. They are formative. They create a kind of internal architecture. The human being that emerges after such effort is not identical to the one that began the task. There is a strengthening involved. The person now sees a little more clearly, acts a little more confidently, and is a little less likely to be deceived by surface level appearances the next time around.
AI changes the economics of this formation. It offers outputs without requiring a commensurate depth of interior transformation. It can often provide the language of understanding without the underlying structure of understanding having been built in the user. This is not always harmful. Sometimes the tool genuinely helps us cross a threshold and see what we could not see before. But in many cases it creates a counterfeit of mastery. One has the appearance of intelligence, the verbal form of competence, and the practical veneer of productivity, without the grounding labour that would ordinarily have made these things one's own.
And when that counterfeit becomes widespread, accountability begins to thin out in curious ways. If the code fails, was it the engineer's failure, or the model's? If the report is shallow, was it the analyst's misunderstanding, or the AI's generic synthesis? If the essay is derivative, was the writer lazy, or merely over-assisted? It is tempting to say that this is irrelevant because the market only cares for results. But markets do care for accountability eventually, especially when systems fail at scale. Someone must still answer for poor reasoning, for fragile software, for wrong decisions, and for the consequences of action taken on shallow understanding. Radical delegation obscures this, because it multiplies outputs while diffusing ownership over the thinking that led to them.
This is why I do not think the future of AI can be properly framed as merely a race between optimists and pessimists. It is a question of anthropology before it is a question of economics. What sort of being does the human become through the regular use of such tools? What habits are reinforced? What forms of laziness are made normal? What kinds of excellence become easier to fake? What kinds of patience become harder to sustain? We ought to be suspicious of any technological regime that makes the visible signs of competence easier to obtain, while making the underlying cultivation of competence optional.
The managerial and institutional form of this problem is easy to imagine. A team discovers that AI lets them ship faster. A manager then asks for more output with the same headcount. Another manager notices that junior engineers can now produce PRs that look acceptable sooner. A process is established where humans review, approve and nudge model outputs at scale. Over time, the institution ceases to reward the slow accumulation of judgment because the visible proxies for judgment are now cheaply available. The system begins to privilege throughput over understanding. A generation of people learns to operate at the communication layer of work while becoming less familiar with the reasoning layer beneath it. This is not merely an issue of preference. It is an issue of what kinds of institutions we are building and what kinds of humans those institutions will select for.
At this point, one could object that every technology has done something similar. Writing changed memory. Calculators changed arithmetic practice. Search engines changed recall. Why should AI be treated differently? My answer is that AI is not merely a storage aid, a retrieval mechanism, or a mechanical accelerator for a well-understood process. AI increasingly intervenes at the level of synthesis, composition, explanation and judgment-like behaviour. It enters the domain that humans once used to become more fully themselves through repeated practice. This is why it cannot be treated as just another convenience. It reaches farther inward.
All of this suggests to me a practical ethic for the use of AI. Use it where leverage is real, but do not use it in a manner that hollows out the very faculties you are hoping to strengthen. Use it to compare approaches after you have attempted one. Use it to critique your reasoning after you have actually reasoned. Use it to accelerate drudgery that does not educate you. Use it to open doors into difficult domains, but do not let it become a permanent substitute for crossing the threshold yourself. In other words, use it in a way that preserves the relationship between effort and growth.
There is a Sanskrit word that feels relevant in this context: dvandva, the realm of dualities and oppositions. Many of our conversations about AI are trapped in precisely such dualities. Optimism versus pessimism. Productivity versus purity. Automation versus authenticity. But perhaps wisdom here lies in a more careful middle path. A refusal of both naive surrender and performative rejection. A deliberate use of tools that keeps the human person in view.
The point, then, is not to ask whether AI can do the work. In many cases, it plainly can. The point is to ask what happens to us when we allow it to do too much of the work that once made us capable, responsible and alive to the world. If the answer is that we become more dependent, less accountable, less skillful, and more alienated from our own effort, then we have not progressed merely because the output arrived faster. Progress has to be evaluated not only by the abundance of artifacts it produces, but by the quality of the humans it leaves behind.