{
  "id": 1853644,
  "title": "The builders of AI are selling a future their own technology destroys",
  "url": "https://urgent.news/2026/08/19/the-builders-of-ai-are-selling-a-future-their-own-technology-destroys",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-19T02:15:22.000Z",
  "source": {
    "name": "e27",
    "slug": "e27",
    "url": "https://e27.co/the-builders-of-ai-are-selling-a-future-their-own-technology-destroys-20260816/"
  },
  "original_language": "en",
  "account": "The construction of artificial intelligence presents a future that its own technology may ultimately undermine. Companies developing AI offer two narratives to employers and workers. Employers are promised AI agents that work continuously at decreasing costs, while workers are told about the rise of the \"agent boss\" - a system where every employee manages a team of AI systems. However, these narratives cannot both represent the ultimate outcome. Supervising human labor has always entailed coordination overhead, including recruitment, training, scheduling, monitoring, and managing morale. However, when work is executed by AI agents, the human approval step becomes a costly element in the workflow, while other expenses continue to decline. This does not imply that humans will disappear from work; rather, the question shifts. Instead of asking \"what can AI do,\" the more pertinent question is \"what must a human still be trusted to approve, sign, represent, or be liable for?\"\n\nEvidence of these shifts is already apparent in the labor market. Researchers from Stanford, analyzing ADP payroll data, discovered that young workers aged 22 to 25 in occupations most exposed to AI experienced a 16 percent relative employment decline from late 2022 to September 2025, compared to less exposed peers. Experienced workers in the same occupations either remained stable or grew. For young software developers, the decline was nearly 20 percent from the late 2022 peak. This decline is not primarily due to mass layoffs but rather firms hiring fewer individuals at the bottom.\n\nThe change is not only visible in the reduction of entry-level positions but also in the shortening of the career ladder. AI systems can perform tasks such as reading, summarizing, drafting, checking, reconciling, searching, formatting, and preparing material for more senior personnel. These tasks, while not useless, are precisely the ones that AI systems can absorb earliest. Consequently, the development of senior judgment is hindered since individuals learn through apprenticeship by performing low-level work and learning from better-performing colleagues. If firms remove the entry rung, they also weaken the pipeline that produces future experts. The irony lies in the fact that those who will be most valuable in the future may have their training path disrupted at present.\n\nTo comprehend this transition, it is essential to separate two curves: the first is machine capability, which evolves continuously as AI systems are produced industrially, improved through software, and made cheaper through scale. The second curve is permission, which moves in steps, allowing machines to perform tasks initially, then with approval, then within guardrails, and eventually autonomously. These steps are primarily about liability, insurance, reversibility, regulation, and whether the other party accepts the machine's actions.\n\nThis gap between machine capability and permission is the cause of much current confusion. Despite AI's ability to perform many tasks, institutions still require human approval, sign-off, certification, explanation, or responsibility. This creates the temporary role of an agent supervisor, which, though temporary, is not artificial. For a period, many workers will indeed manage AI tools, specifying goals, checking outputs, designing workflows, and approving actions. Microsoft's \"agent boss\" framing aptly captures this transitional phase. However, the same cost logic that creates the supervisor role also applies pressure to it. As AI workflows run continuously, and every machine step becomes cheaper, the human approval gate becomes an increasingly significant portion of the total cost.\n\nThe solution is not for humans to work 24 hours a day to keep up with machines. Rather, machines are being promoted up the trust ladder. Career moats, which are durable advantages that protect a business from competition, also play a role in this transition. A friction-based moat arises from tasks being slow, opaque, tedious, or difficult to coordinate, such as being faster at reading documents or managing process throughput. However, these advantages are often task properties that AI can quickly substitute. A structural moat, on the other hand, exists because the value depends on factors machines cannot cheaply inherit, such as legal liability, physical presence, social legitimacy, elected or appointed authority, trusted representation, or judgment in domains with no clear right answer. The labor market has yet to differentiate these two types of moats effectively. A résumé does not indicate whether a person's value stems from scarce judgment or merely from friction that used to be expensive to remove. This distinction is now becoming crucial, as it pertains to the value a person can bring in an AI-dominated world.",
  "summary": "Every salary is a bet that something about its holder stays valuable. AI is changing the terms of that bet. The companies building the AI future are selling that future in two different ways. To employers, they promise agents that work around the clock at falling cost. To workers, they promise a promotion: the age […] The post The builders of AI are selling a future their own technology destroys…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}