AI-Native Firms, Legacy Illusions, and the Time Lag People Are Not Seeing
The Comforting Story That Is Wrong
Over the last year, a new kind of article has started to appear. It does not portray artificial intelligence as an existential threat, nor as a miracle cure. Instead, it focuses on disappointment. Executives report that AI initiatives have underperformed expectations. Pilot projects have struggled to move beyond experimentation. Organizations that anticipated rapid transformation often find themselves wrestling with integration challenges, governance concerns, data quality issues, and uncertain returns on investment (McKinsey & Company, 2024; Maslej et al., 2025).
At the same time, a parallel narrative has emerged around the idea that the most valuable AI skill is learning how to communicate effectively with the technology. According to this view, those who master prompt-writing will possess a durable competitive advantage. The implication is that the future belongs to “AI whisperers” who know the right words to unlock the technology’s potential and thereby secure their professional relevance.
These narratives often end on a reassuring note. They emphasize the enduring value of human judgment, creativity, and empathy. They envision a future in which artificial intelligence handles routine tasks while human beings continue to perform the meaningful work. Productivity increases, jobs become enriched, and organizations achieve a stable balance between human and machine capabilities. For workers concerned about their future employability, it is an attractive and comforting story.
The quiet assumption underlying this narrative is that what we see today is roughly what AI is—and what it will remain. If that assumption were true, then perhaps a modest investment in prompt-writing skills and some incremental workflow adjustments would indeed be sufficient. However, that is not what current evidence suggests.
The systems now entering workplaces are improving on a timescale measured in months rather than decades. Recent advances have produced substantial improvements in reasoning capabilities, tool use, long-context processing, multimodal performance, coding ability, and structured output generation (Maslej et al., 2025; McKinsey & Company, 2024). While many organizations continue to experience frustration with implementation, the underlying technology continues to advance rapidly.
What appears from the outside to be a somewhat unreliable assistant is increasingly becoming something more significant: a new kind of worker. That distinction matters because the central issue is not whether a chatbot impressed a manager this quarter. The real question is whether these capabilities are already strong enough to be embedded within organizations whose economic structures differ fundamentally from those most people work in today.
What Incumbents Are Discovering the Hard Way
Inside large organizations, artificial intelligence usually arrives as an addition rather than a redesign. It appears as a chatbot on the intranet, an email assistant, a meeting summarization tool, or an automated layer attached to an existing process. From this perspective, reports of disappointment are understandable because organizations encounter the limitations first.
They encounter systems that occasionally generate inaccurate outputs. They discover workflows that break down in edge cases. They face difficulties integrating AI into legacy software systems, fragmented databases, and established governance structures. Compliance requirements, security concerns, and organizational politics frequently slow implementation efforts (McKinsey & Company, 2024; Maslej et al., 2025).
Many executives have consequently reframed their ambitions. Instead of pursuing large-scale transformation, they often position AI as a tool that supports employees rather than fundamentally altering organizational structures. This reflects not only caution but also the reality that many organizations were designed around human workers performing cognitive tasks within processes developed long before modern AI existed.
What these organizations are discovering, however, may reveal as much about the limitations of incumbent firms as about the limitations of AI itself. Their data systems are often fragmented. Their workflows evolved around human decision-making. Their governance structures assume that information moves at human speed. As a result, the first generation of enterprise AI deployments has frequently produced incremental efficiencies rather than transformational change (McKinsey & Company, 2024).
This distinction is important. What appears to be a limitation of AI inside a legacy organization may actually be a limitation of the organization itself. A technology that struggles when inserted into an existing structure may perform very differently when the structure is designed around the technology from the beginning.
That observation leads to a more important question: What happens when a company is built from the ground up on the assumption that language models and AI agents are not merely assistants, but core workers?
An AI-native company is not an old business with some artificial intelligence sprinkled on top. It is a business whose structure assumes that current-generation language models and agents can already perform a substantial share of routine cognitive work. Tasks such as gathering information, processing requests, applying standard rules, drafting routine communications, answering common questions, and moving work through defined workflows can increasingly be performed by AI systems operating under human supervision (Brynjolfsson et al., 2025; McKinsey & Company, 2024; Maslej et al., 2025).
In such a company, human beings do not disappear, but their role changes.
Rather than spending large portions of their time executing routine processes, human workers increasingly design workflows, define decision thresholds, monitor performance, handle exceptions, build client relationships, and exercise judgment where accountability, trust, and ambiguity matter. The organization becomes less dependent on large numbers of people performing repetitive cognitive work and more dependent on smaller numbers of people overseeing systems that perform much of that work automatically (Brynjolfsson et al., 2025).
Importantly, this model does not depend on the assumption that AI is perfect. It depends on a much more realistic observation: neither human beings nor machines are perfect. Every organization already operates with non-zero error rates. Payroll departments make mistakes. Customer service representatives provide inconsistent answers. Administrative staff overlook details. Managers occasionally make poor decisions. Perfection has never been the standard by which organizations operate. Instead, organizations seek acceptable performance at acceptable cost.
This distinction is critical because AI is often judged against an implicit standard of perfection while human performance is judged against a standard of familiarity. Human errors are expected and therefore frequently tolerated. AI errors are novel and therefore attract disproportionate attention. Yet markets rarely reward perfection. They reward combinations of quality, speed, reliability, and price that customers find acceptable.
An AI-native organization is built around this economic reality. Data is stored in machine-readable formats. Workflows are designed for automation from the outset. Governance mechanisms are embedded within processes rather than layered on afterward. Escalation pathways, audit trails, confidence thresholds, permissions, and quality controls are incorporated into the operating model itself (McKinsey & Company, 2024).
Perhaps most importantly, none of this requires a fundamentally new form of artificial intelligence. Much of the recent progress has emerged from continued improvements within the existing large-language-model paradigm. Better reasoning, enhanced tool use, retrieval systems, agent orchestration, longer context windows, and more effective post-training methods continue to expand the range of tasks these systems can perform (Maslej et al., 2025).
Once that structural shift becomes visible, the next question is economic. If firms can be organized around these assumptions, what happens to cost, pricing, and competitive advantage?
Why Cost and Speed Tilt Toward the New Players
The real difference between a legacy firm and an AI-native one is structural cost.
Most established service organizations carry significant labor costs. They employ administrators, coordinators, analysts, support personnel, supervisors, managers, and specialists. These employees are supported by offices, software platforms, training programs, compliance structures, and layers of management. All of these costs are embedded within the final price paid by customers.
An AI-native competitor may still require human expertise, but it often requires fewer people to deliver the same routine service. Human effort becomes concentrated on exceptions, oversight, client relationships, and strategic decisions, while routine work is handled through automated workflows supported by AI systems (Brynjolfsson et al., 2025).
Take payroll as an example. A traditional provider might need to charge something like 13,000 dollars per client each year simply to carry its wage bill, overhead, and profit expectations. An AI-native rival with a lean staff and a workflow built around agents could potentially offer the same basic service for 8,000 dollars, or even 4,000 to 5,000 dollars for a time if it were willing to live on thinner margins while it won customers. Phase one is not maximizing profit. Phase one is making the legacy price structure look unreasonable.
Customer support follows the same logic. Research has demonstrated that generative AI can substantially increase productivity in customer-support environments while improving quality, particularly among less experienced workers (Brynjolfsson et al., 2025). If a significant proportion of routine interactions can be handled through AI-assisted processes, organizations may be able to provide comparable service using fewer personnel while maintaining acceptable quality levels.
The same logic applies across a wide range of information-processing functions. Scheduling, document preparation, records management, compliance monitoring, onboarding, benefits administration, technical support, and many other activities consist largely of moving information through predictable workflows. To the extent that those workflows can be standardized, they become candidates for increasing levels of automation.
Scale compounds the advantage.
Traditional firms generally grow by hiring additional employees, training them, supervising them, and integrating them into existing structures. Growth is constrained by labor availability, training capacity, management overhead, and organizational complexity.
AI-native firms face different constraints. They still require human expertise, but growth increasingly depends on infrastructure, data integration, workflow design, and computational resources rather than proportional increases in headcount. Additional demand can often be accommodated by extending systems that already exist rather than building entirely new teams (McKinsey & Company, 2024; Maslej et al., 2025).
This does not imply frictionless growth. Infrastructure costs remain significant. Trust must still be earned. Regulations must still be followed. Data integration remains difficult. Human oversight remains necessary. Nevertheless, the nature of the bottleneck changes. The limiting factor becomes less about recruiting and managing large numbers of workers and more about building and maintaining effective systems.
That distinction matters because competitive markets rarely care how incumbents are organized. Customers generally respond to combinations of price, quality, reliability, and convenience. If a competitor can deliver acceptable service at significantly lower cost, market pressure eventually follows.
This is why the important question is not whether today’s AI pilots are disappointing. The more important question is what happens when organizations built around fundamentally different cost structures begin competing directly with traditional firms.
The Time Lag That Hides What Is Coming
If these shifts are real, why do so many people still feel that nothing has changed?
Part of the answer lies in organizational inertia. Large institutions are remarkably effective at absorbing change. Regulations, contracts, budgeting cycles, governance processes, legacy systems, union agreements, and cultural norms all slow the visible pace of transformation. Even when technological capabilities improve rapidly, organizational change often unfolds much more slowly (McKinsey & Company, 2024).
Another part of the answer comes from the history of innovation itself.
Technological adoption rarely follows a straight line. Instead, innovations typically diffuse through populations in an S-shaped pattern. Early progress appears slow and uncertain. Adoption remains concentrated among innovators and early adopters. Then, once certain thresholds are crossed, diffusion accelerates rapidly and what previously appeared marginal becomes mainstream (Rogers, 2003).
This pattern has been observed repeatedly across technologies, industries, and historical periods. Moore (1991) described the challenge of moving from early adopters to mainstream markets as “crossing the chasm.” Christensen (1997) demonstrated how disruptive technologies often appear inferior to incumbents in their early stages, only to improve rapidly and eventually reshape entire industries.
Viewed through this lens, the current state of enterprise AI becomes easier to understand.
Many organizations remain stuck in the difficult middle stage. They are attempting to layer new technologies onto old processes. They are experimenting without redesigning the work itself. Consequently, progress often appears slower and less dramatic than advocates predicted.
At the same time, evidence suggests that AI adoption, investment, and deployment continue to expand rapidly across industries (Maslej et al., 2025). This combination—a slow experience inside many organizations coupled with rapid technological improvement and growing investment—is precisely what periods of technological transition often look like.
Most people experience continuity while underlying economic structures begin to change.
That is why the risk often feels abstract until it suddenly does not. By the time employees encounter the consequences directly—in restructurings, outsourcing decisions, workforce reductions, or major process redesigns—the conditions that made those changes possible may have been developing for years.
The gap between technological capability and personal experience is where false reassurance thrives.
What This Means if You Work in an Established Organization
From inside a large organization, the slow pace of visible change can feel reassuring. Artificial intelligence often appears unreliable, inconsistent, and overhyped. Pilot projects stall. Managers hesitate. Existing systems continue to operate much as they always have. Daily work remains recognizably familiar.
That experience is real. However, it may also be misleading.
The threat facing many workers does not arise solely from their employer directly automating their role. It may emerge from a more indirect source: competition. Organizations increasingly compete not only on the quality of their services but also on the cost structures that support those services. If AI-native firms can deliver comparable outcomes at lower cost, the pressure eventually reaches incumbent organizations whether they embrace automation aggressively or not (Brynjolfsson et al., 2025; Christensen, 1997).
A payroll department may not disappear because internal AI systems suddenly become perfect. Instead, the department may face growing pressure if external providers can deliver acceptable payroll services at substantially lower cost. A customer-support team may not vanish because a chatbot achieves flawless performance. It may shrink because another organization can provide satisfactory support with a leaner operating model.
The distinction matters.
Workers often focus on whether AI can completely replace their specific job. The more important question may be whether AI changes the economics of delivering the service in which that job exists. Market pressures do not require perfect automation. They only require sufficient improvements in cost, speed, scalability, or convenience to create a competitive advantage.
This is why organizational inertia should not be confused with long-term protection. Large institutions often move slowly, and that slowness can temporarily shield employees from immediate disruption. However, it tells us relatively little about what becomes possible when organizations redesign workflows, restructure operating models, and adapt to changing economic realities (McKinsey & Company, 2024).
The roles most exposed are often those that sit in the thick middle of information work: positions built around processing information, applying established rules, producing standardized outputs, coordinating routine workflows, and moving requests from one stage of a process to another. These activities have historically required large numbers of human workers. Increasingly, they are becoming candidates for varying degrees of automation and augmentation.
This does not mean human beings become irrelevant.
In fact, the opposite may be true.
As routine work becomes easier to automate, the relative value of distinctly human capabilities may increase. Judgment becomes more important when situations are ambiguous. Trust becomes more important when accountability matters. Relationships become more important when decisions affect people. Adaptability becomes more important when circumstances change faster than rules can be written.
For workers, the lesson is not that prompt-writing will become a magic profession. Prompt engineering may be useful today, but history suggests that interfaces generally become easier over time. Few people today build careers around memorizing operating-system commands or programming printer drivers. Technologies mature by becoming easier to use, not harder.
The more durable advantage lies elsewhere. It lies in developing the ability to understand complex situations, exercise sound judgment, build trust, solve novel problems, and adapt to changing conditions. Those capabilities are difficult to automate precisely because they depend on context, experience, relationships, and an understanding of what matters when the rules no longer fit neatly.
In an economy increasingly influenced by AI-native organizations, these are not merely desirable personal qualities. They are increasingly valuable economic assets.
The danger in the current moment is not simply that people underestimate artificial intelligence. It is that many misunderstand the form in which its economic impact is likely to arrive.
Some observers look at disappointing pilot projects within large organizations and conclude that AI has stalled. Others embrace the comforting belief that learning to write better prompts will be enough to secure their professional future. Both interpretations may underestimate the scale of the structural changes that are already underway.
The current generation of language models has demonstrated capabilities that are increasingly sufficient to support organizations built around fundamentally different assumptions about labor, workflow design, scalability, and cost structure (Brynjolfsson et al., 2025; Maslej et al., 2025). These organizations do not require perfect systems to become powerful competitors. They require systems that are reliable enough, affordable enough, and improving rapidly enough to create economic advantages that incumbent firms struggle to match.
History suggests that disruptive technologies rarely transform industries overnight. Instead, change often unfolds gradually, hidden beneath the surface of everyday experience. Established organizations continue to function. Existing jobs remain in place. Most people experience continuity. Yet beneath that apparent stability, competitive pressures slowly reshape the economic landscape until the consequences become impossible to ignore (Christensen, 1997; Rogers, 2003).
For ordinary working people, this is ultimately not a story about artificial intelligence laboratories, venture capital investments, or software architectures. It is a story about livelihoods.
It is about the parent trying to keep a mortgage paid. The mid-career professional wondering whether decades of experience still matter. The worker who followed every rule, acquired every credential, and now finds the nature of work changing beneath their feet.
The comforting stories of the present suggest that little will change or that a new technical skill will provide lasting protection. The evidence points toward a more challenging reality. The most durable response is unlikely to be mastering a particular prompt format or learning the latest AI tool. It is developing the judgment, adaptability, critical thinking, relationship skills, and depth of understanding that become more valuable precisely because so much routine work can increasingly be performed by machines.
The future may belong neither to the machines nor to the people who know the best prompts.
It will belong to those who best understand how to exercise human judgment in a world increasingly shaped by intelligent systems.
Brynjolfsson, E., Li, D., & Raymond, L. R. (2025). Generative AI at work. Quarterly Journal of Economics, 140(2), 889–943. https://doi.org/10.1093/qje/qjae044
Christensen, C. M. (1997). The innovator’s dilemma: When new technologies cause great firms to fail. Harvard Business School Press.
Maslej, N., Fattorini, L., Perrault, R., Gil, Y., Parli, V., Kariuki, N., Capstick, E., Reuel, A., Brynjolfsson, E., Etchemendy, J., Ligett, K., Lyons, T., Manyika, J., Niebles, J. C., Shoham, Y., Wald, R., Walsh, T., & AI Index Steering Committee. (2025). Artificial Intelligence Index Report 2025. Stanford Institute for Human-Centered Artificial Intelligence. https://hai.stanford.edu/ai-index
McKinsey & Company. (2024). The state of AI in early 2024: Gen AI adoption spikes and starts to generate value. McKinsey Global Survey on AI. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-2024
Moore, G. A. (1991). Crossing the chasm: Marketing and selling disruptive products to mainstream customers. Harper Business.
Rogers, E. M. (2003). Diffusion of innovations (5th ed.). Free Press.