“There will be a lot of trauma and disruption along the way,” Elon Musk told Joe Rogan, grasping at an analogy that resonates with anyone who observes the process of AI gradually eating into professional routines. He termed it a “supersonic tsunami,” a wave that does not come with gradual accumulation of the previous workplace technology but with the disturbing pace of software upgrades.
The most pristine distinction in the framing of Musk is not between industries but two types of work: data manipulation work and atoms manipulation work. Generative AI has already broken the barrier where it can write, summarize, search, categorize, and rephrase in large scale, and thus makes the jobs that can be performed behind the screen particularly vulnerable. The tangible economy continues to play the role of a decelerator. Edges, safety, and dexterity issues that involve cooking, construction, and maintenance of fields are replete with edge cases, safety limits, are not being broadly solved by current robotics, despite the improvements in prototypes.
The ability gap is now visible on the office side. Big language models like Claude, ChatGPT, and Gemini are capable of writing and debugging code, synthesizing documents, generating analyses, and following multi-step directions. Dario Amodei, the CEO of Anthropic, has contended that the law, consulting, administration and finance have repetitive but variable work that is susceptible in the short term, observing that AI systems are already quite capable of handling core desk work, and are rapidly improving in doing so. The caution issued by Musk falls on the same sentence, with digitally mediated thinking being not only cheap, fast, and more reliable, but the economic logic of substitution makes it difficult to overlook by employers due to the collapse of governance barriers.
The fact that no humanoid robots or new factories are needed to accomplish the displacement is one of the reasons why the “tsunami” metaphor is still being used. It operates on the current IT budgets and comes as a service. Within a world of spreadsheets and ticketing systems, it is possible to automate workflow one workflow at a time: write the first reply, create the weekly report, create the first contract markup, etc. all the way up until the job looks less like work and more like an audit of a job. This change is important since the structural labor demand of auditing is lower than the structural labor demand of producing especially when the output quality can be driven high through repetition. It also transforms the people who are hired. Once an organization is able to purchase competence as a subscription, entry-level jobs formerly used as training ramps become the simplest to compress, with fewer access points to white-collar jobs available in the first place before economy-wide statistics can be seen to shift visibly.
Measurements of potential exposure show how wide the blast radius could be. OpenAI’s occupation-based metrics suggest that 30% of workers could see at least half of their tasks disrupted by generative AI, with office and administrative support among the most exposed categories. That does not mean jobs disappear in a single step, but it does signal that many roles can be redesigned around machines performing the first pass and humans handling exceptions, approvals, and accountability.
Ironically, businesses have begun to experiment with that redesign in front of the audience. By 2025, AI was listed on the list of restructurings and cost-cutting in major employers, and Challenger, Gray and Christmas predicted nearly 55,000 U.S. layoffs this year as a result of AI. Beth Galetti, an employee of Amazon, said that generative AI was “the most undergoing technology since the Internet,” explaining the reason the company wanted to become more leanly organized. Salesforce CEO Marc Benioff put the mechanism in simple words: “I have cut it down to 9,000 heads to about 5,000, because I require fewer heads.” Other executives also talked about job mix moves and not just cuts. According to IBM CEO Arvind Krishna, AI chatbots had replaced the jobs of several hundred HR employees as the company added more positions in other fields that demand greater critical thinking.
Meanwhile, the macroeconomic image is not cinematic enough in comparison with the metaphors. Scholars of Yale, in the Budget Lab, analyzing the occupational composition since the launch of ChatGPT, say that the labor market is more stable than they had expected to collapse, and that the ratio of employment in the exposure groups has been exceptionally stable. Their rationale is procedural and not philosophical: even transformative technologies need time to spread: governance, complementary investment, and process change is needed to enable adoption. The same analysis reveals a major mismatch that technology vendors seldom focus on high theoretical exposure does not necessarily mean high usage and usage does not necessarily mean automation.
The next stage is being constructed as that divide between the models and what organizations are actually implementing. The adoption in the workplace was initially biased towards augmentation: employees were using chat tools to write drafts, brainstorm and accelerate normal job processes, since it can be comfortably placed within current job roles. The risk to human count increases as companies cease to sprinkler chatbots into work and rather re-engineer workflows on the basis of “agentic” systems, which perform end-to-end: intake, retrieval, drafting, routing, updating systems of record, and closing loops with little human intervention.
The employment statistics at Indeed are an indication that the transition is a skills issue, rather than a jobs issue. Its GenAI Skill Transformation Index concludes that 26% of jobs recently posted on the platform can be significantly transformed with the assistance of generative AI and that 46% of the skills of a typical posting are in a grey area where AI can perform much of the routine tasks but humans are in charge of supervising. The implication is structural: most jobs do not go away, but the human aspect shifts to verification and exception-handling and to stakeholder judgment and compliance-work, which is narrower, more senior and more difficult to fill through standard pipelines.
The longer-run refutation by Musk is that there would be abundance after the disruption. He has explained a “benign situation” whereby AI and robotics allow the realization of universal high income, making work optional since output is so cheap and abundant. The humanoid robot project, Optimus, by Tesla, fits that story as the transition between the software productivity and the physical productivity. However even optimistic accounts of embodied AI still run into a practical limitation that Musk himself points to implicitly, namely abundance: automation must get off the screen and safely scale in untidy situations, not just in demonstrations.
What makes the present moment feel unstable is the mismatch in tempos. Model capability and tool access iterate quickly, while institutions—hiring systems, professional credentialing, liability rules, and internal controls—move slowly. The result is a workplace that can look stable in aggregate data while still feeling turbulent within specific teams, especially in roles where output is already digital and easy to benchmark.
The “supersonic tsunami” framing persists because it captures that asymmetry: the shoreline is not the economy as a whole, but the subset of work that can be turned into standardized, auditable tokens of text, code, and analysis. Once that conversion happens, the question becomes less about whether AI can do the work and more about how quickly organizations decide they trust it to run the workflow.
