The Broadband Moment: Why Gen AI Proficiency No Longer Makes You Exceptional

image courtesy of Gen AI - prompt author

Introduction

There is a particular kind of unease that settles over a profession when a tool once considered a mark of distinction becomes table stakes. Ask any consultant who once billed a premium for "digital transformation expertise" in 2015, or any analyst who charged extra for "advanced Excel modelling" in 1998, and they will describe the same arc: capability becomes commodity, commodity becomes infrastructure, and infrastructure becomes invisible. We are watching this arc complete itself again, this time with generative artificial intelligence, and the speed of the transition should give scholars and consultants alike considerable pause.

For the past three years, the professional conversation around Gen AI has centred on acquisition of skill: how to prompt effectively, how to engineer context, how to chain tools together, how to coax a large language model into producing something useful rather than merely plausible. This was, for a time, a genuine differentiator. Those who understood the craft of prompting produced visibly better outputs than those who did not. But that era is closing, and it is closing faster than most professional service firms have adjusted their pricing models, their training programs, or their sense of what makes their people valuable.

This essay argues that Gen AI has completed its transition from a scarce professional advantage into ambient infrastructure - comparable in kind, if not in speed, to the broadband transition of the early 2000s. It further argues that this transition creates a specific and uncomfortable paradox for knowledge workers: proficiency with the tools no longer explains superior performance, because everyone now has access to comparable proficiency. What remains as a source of differentiation is a narrower and more human set of capacities - taste, domain context, distribution, and trust- that Gen AI does not manufacture and cannot substitute for. Understanding this shift is not an academic exercise. It is a question of professional survival.

The Infrastructure Analogy: What Broadband Actually Taught Us

It is worth dwelling on the broadband comparison rather than treating it as a passing rhetorical flourish, because the historical parallel is more precise than it first appears.

In the early 2000s, a fast, reliable internet connection was a genuine competitive asset. A firm with broadband could move data, host richer client interactions, and operate at a pace that dial-up competitors could not match. Access was unevenly distributed, expensive relative to income, and geographically constrained. Having it conferred advantage. Not having it was merely a limitation, not yet a disqualification. Within a decade, that asymmetry inverted entirely. Broadband did not remain a differentiator; it became a precondition for participating in commercial life at all. No one today wins a client engagement by mentioning they have a reliable internet connection. The absence of one, however, would end the conversation before it began. The advantage evaporated not because broadband became less useful, but because it became universally available - and universal availability collapses the distance between the well-resourced and the merely adequate.

Gen AI is tracing an almost identical curve, only compressed into a fraction of the time. What took broadband roughly a decade to accomplish, large language models appear to be accomplishing in two to three years. The tools have moved from frontier laboratories to consumer applications to embedded features inside the ordinary software that scholars and consultants already use daily - word processors, presentation software, research databases, even email clients. The result is the same structural inversion: possessing Gen AI capability was briefly an advantage; lacking it is rapidly becoming a liability; and soon, possessing it will be assumed, the way electricity, telephony, and broadband are assumed.

This has a specific implication that deserves to be stated plainly, because it cuts against much of the current professional discourse: if everyone can produce competent work with Gen AI, competent work is no longer the currency of competitive advantage. The bar for adequacy has risen sharply. The bar for exceptional has not moved nearly as much, and in some domains, it may not move at all, because exceptional work has never been primarily a function of tool access.

The Conundrum of Universal Access

Here is where the genuine intellectual puzzle sits, and it is worth stating in its sharpest form rather than softening it. For several years, the professional literature on Gen AI - the webinars, the LinkedIn essays, the internal training decks circulated by consultancies - has insisted that mastery of prompting, of context engineering, of retrieval strategies, and of tool orchestration constitutes a durable professional skill. And for a time, this was true. Early adopters who understood how to structure a prompt, supply the right context, and iterate toward a usable output produced work that was measurably superior to that of their peers who treated the tools as novelties.

But skills that can be documented, taught, and templated are, by their nature, skills that diffuse quickly. Prompting technique is not tacit knowledge locked inside years of apprenticeship; it is closer to a recipe. Recipes travel. They travel through blog posts, through corporate training programs, through the models themselves, which have become steadily better at inferring intent from underspecified instructions - effectively absorbing the burden of good prompting into the system itself, so that the human need no longer supply it. The very thing that made prompting valuable - that it could be learned and transmitted - is the thing guaranteeing its evaporation as a differentiator. This produces the conundrum the introduction gestures toward: if the skill that was supposed to separate the sophisticated user from the casual one is now accessible to everyone with a browser, on what basis does anyone claim superior work? The honest answer is uncomfortable for an industry that has built entire practice areas around "AI enablement" and "prompt engineering" training: the basis cannot be the tool use itself. It must be something upstream or downstream of the tool - something the tool does not provide and cannot be trained into existence through a better prompt.

A Case in Point: Stitch, Flow, and the Collapsing List of Exceptions

Sceptics of this argument have historically pointed to specific domains as evidence that Gen AI's reach has limits - visual design, video production, the translation of a rough concept into a finished creative asset. These were the last redoubts, the areas where human craft was assumed to retain an irreducible advantage because the work required taste and technical execution simultaneously. That list of exceptions is shortening in real time. Google's Stitch, for instance, can move from a simple sketch to a high-resolution mock-up, and can take that mock-up further still - generating a functioning website, populating it with contextually appropriate copy, and handling the kind of implementation detail that once required a small team of designers and developers working in sequence. Google's Flow performs an analogous function for video, producing footage usable in real commercial contexts rather than mere demonstration reels. Neither tool requires the user to possess deep technical fluency in design software or video editing. Both compress what was once a multi-role production pipeline into a single conversational interface.

The natural question - the one the introduction poses directly - is what remains that Gen AI cannot produce adequately. It is a fair question, and scholars in particular should resist the temptation to answer it with a comfortable platitude about human creativity being irreplaceable. The more rigorous answer is narrower and less flattering: Gen AI can now produce adequate output across a widening range of tasks that once required specialised professional training. Adequate is no longer the threshold that separates the professional from the amateur, because the tools have made adequate available to both. This returns us to the broadband hypothesis in its most literal form: possessing this capability does not make an individual or a firm distinctive. It merely ensures they remain in contention. Its absence, increasingly, ensures they do not.

The Quality Floor and the Unmoved Ceiling

It is worth pausing on a distinction that is easy to state and consequential in its implications: Gen AI has raised the floor of acceptable output without correspondingly raising the ceiling of exceptional output. Consider what the internet looked like before generative tools became widespread. It was, by broad consensus, uneven - a landscape in which genuinely excellent work sat alongside a great deal of content that was poorly researched, badly structured, or simply careless. Distinguishing the excellent from the mediocre was, in a strange way, easier then, precisely because mediocrity was often visibly mediocre. A reader or client could tell within moments whether they were looking at serious work or a hastily assembled substitute.

That visible gradient has narrowed. Gen AI-produced content, competently prompted, rarely reads as careless. It is grammatically sound, structurally coherent, reasonably well-researched within the limits of its training and retrieval, and largely free of the obvious markers that once signalled low effort. The result is a landscape saturated not with poor content but with competent, interchangeable content - work that is difficult to fault on any individual criterion and equally difficult to remember five minutes after encountering it. This is a harder environment to compete in than the one it replaced, not an easier one, and this point deserves emphasis because it runs counter to much of the popular narrative around Gen AI as a great leveller. A rising floor without a rising ceiling means the competitive advantage once available to anyone who simply cleared a modest bar of quality no longer exists. Everyone clears the modest bar now. What separates a Financial Times op-ed from a passable LinkedIn post is no longer that one is well-written and the other is not; both may be well-written. It is something else entirely - a set of qualities that generation, however sophisticated, does not manufacture on its own. Scholars and consultants, whose professional value has long rested on the claim of judgment above execution, should find this observation clarifying rather than alarming. It restates, in sharper terms, what has always been true of expertise: execution was never the scarce resource. Judgment was.

The Four Differentiators

If tool proficiency has become a prerequisite rather than an advantage, the question of what now constitutes genuine differentiation becomes the central professional question of this decade. Four attributes stand out as durable, and each merits individual treatment.

Taste

Taste is the capacity to recognise, among many technically adequate options, which one is actually right - for this audience, this moment, this argument. Gen AI can generate variations at essentially no marginal cost; it cannot reliably select among them with judgment that accounts for context the model was never given and cannot infer. Taste is what allows a consultant to look at five competent slide decks generated in minutes and know that all five are wrong for the client in the room, or to recognise the one phrase in a report that will land badly with a particular board. This is not a mystical or ineffable quality. It is accumulated pattern recognition, built through exposure to consequences over years of practice, and it remains stubbornly human because it depends on lived professional experience the model does not have and cannot borrow.

Domain Context

Generic competence is now abundant; specific, situated understanding is not. A model can produce a plausible analysis of a merger, a policy proposal, or a market entry strategy, but it does so without knowledge of the particular history between two negotiating parties, the unstated political constraints inside a client organisation, or the specific failure of a similar initiative eighteen months earlier that no one has written down anywhere the model could retrieve it. Domain context is the accumulated, often undocumented understanding of how a particular field, institution, or set of relationships actually functions beneath its formal description. It is precisely the kind of knowledge that resists generalised training data because it is local, current, and frequently confidential.

Distribution

Producing excellent work has never been sufficient on its own; ensuring the right people encounter it has always been a separate and equally demanding discipline. As the volume of competent Gen AI-assisted content multiplies, the bottleneck shifts further downstream, from creation toward reach. A scholar with an original insight and no channel to communicate it competes poorly against a mediocre insight distributed through an established network, a trusted newsletter, or a relationship built over years. Distribution - the audience, platform, and relational capital that carries work to the people who matter - has always mattered, but a saturated content environment makes it decisive rather than merely helpful.

Trust

Finally, trust: the accumulated confidence that a particular person's judgment can be relied upon, especially in circumstances where the underlying work cannot be easily verified in the moment. Trust is not manufactured by output quality alone. It is built through a track record, through the visible bearing of accountability when things go wrong, and through relationships that predate any single deliverable. A client does not merely want a good report; they want the assurance that the person standing behind the report has judgment they can rely on when the report is wrong, incomplete, or politically inconvenient. No model can offer that assurance, because no model bears consequences.

Conclusion: Competing After the Floor Has Risen

Everyone now uses Gen AI; now what? The broadband analogy, taken seriously, offers both a warning and a form of relief. The warning is straightforward: firms and individuals who continue to market Gen AI fluency itself as their primary value proposition are building a strategy on a foundation that is actively dissolving beneath them, much as firms that once marketed "we have high-speed internet" would appear faintly absurd today. The relief is that this dissolution clarifies rather than eliminates the basis for professional distinction. It was never really about the tool.

Scholars and consultants operating in this environment would do well to treat Gen AI exactly as the broadband analogy suggests: as necessary, unremarkable infrastructure to be adopted quickly and without ceremony, freeing attention for the harder and more durable work of cultivating taste, deepening domain context, building genuine distribution, and earning trust over time. These four attributes were always the substance of expertise. Gen AI has simply stripped away the layer of technical execution that used to obscure how much of professional value rested on them all along.