The AI Jobs Future Will Be Uneven, Not Empty

The AI jobs future will probably be uneven: fewer routine entry routes, more work done with software assistance, and a long argument over who receives the productivity gains. Dylan John’s video “MIT Just Discovered What Jobs AI Actually Killed” makes that case through a useful contrast between a sudden “crashing wave” and a gradual “rising tide.” The evidence suggests that both possibilities deserve attention, but neither should be treated as a settled forecast.

The future may arrive first at the bottom of the ladder

The video’s most important idea is not that every job disappears. It is that the first visible damage may happen to the jobs people use to enter a profession. Junior analysts, assistant writers, entry-level programmers, translators, and customer-support workers often perform structured tasks under supervision. Those tasks are precisely where a language model can be useful: summarize a document, draft an answer, translate a passage, find a pattern, produce a first version of code, or classify a request.

That creates a difficult possibility. A company may still need experienced people who can set direction, make judgments, handle exceptions, and take responsibility. It may simply need fewer beginners to do the preparatory work. The result would not look like an occupation vanishing overnight. It would look like the first rung of the ladder becoming narrower.

Dario Amodei’s much-discussed warning illustrates the scale of the fear. As Axios reported, Anthropic’s CEO said AI could eliminate half of entry-level white-collar jobs and push unemployment to 10–20% within one to five years. That is a serious warning from a person close to the technology, but it remains a forecast. It is not evidence that the outcome is inevitable, and it should not be confused with a measured estimate of jobs already lost.

The distinction matters because early labor-market data look less dramatic at the economy-wide level. The Budget Lab at Yale found no discernible broad disruption in the first 33 months after ChatGPT’s release. Its analysis also found no relationship, so far, between measures of AI exposure, automation, and augmentation and changes in employment or unemployment. That does not disprove a future shock. It does show why headlines about the future should not be written as though they describe the present.

Crashing waves and rising tides

A crashing wave is the video’s name for a technical jump that suddenly makes a long, difficult task practical. If a system could reliably complete a process that once took a team several days, a company might reorganize around it quickly. Some jobs would be exposed not because the whole occupation is easy, but because one bottleneck task has suddenly become cheap.

A rising tide is slower. Tools improve, workers learn to use them, managers redesign processes, customers adapt, and regulation catches up. In that scenario, the number of tasks that require a human changes gradually. People who once spent their day producing a first draft may spend more time checking, editing, interviewing, deciding, or persuading. Some positions shrink, but the occupation changes shape rather than disappearing at once.

Evidence from workplaces supports the idea that adaptation is possible, although it does not guarantee a gentle transition. The OECD’s survey of employers and workers found generally positive views of AI’s effect on performance and working conditions, alongside real concerns about job loss. It also found that training and worker consultation were associated with better outcomes. In plain language, people do better when they are not simply handed a new tool and told to absorb the consequences alone.

The likely future is therefore a series of local waves inside a broader rising tide. One department may automate a task quickly while another uses AI as an assistant. One occupation may lose routine entry work while another expands because lower costs create new demand. The national statistics can look calm while specific groups experience a genuine crisis.

Translation will not vanish, but routine translation is under pressure

Dylan John presents translation as an example of a crashing-wave occupation. The underlying point is credible: if a customer needs a rough internal translation, a machine can produce a usable first pass immediately. That can reduce the amount of paid work available for routine, low-context assignments.

But “translation is obsolete” is stronger than the evidence supports. The Bureau of Labor Statistics outlook for interpreters and translators projects 2% employment growth from 2025 to 2035, slower than average, with about 6,000 openings per year. Those openings are mostly replacement demand, not a boom. The picture is pressure and limited growth, not extinction.

The surviving value will increasingly sit in places where mistakes are expensive or context is difficult. Legal, medical, diplomatic, literary, and highly technical work needs judgment about meaning, confidentiality, tone, and consequences. A machine can draft a translation; a person may still need to determine whether the result is safe to send, culturally appropriate, contractually faithful, or comprehensible to the intended audience.

For ordinary workers, the warning is practical. “I translate words” is a weaker position than “I help a particular kind of customer communicate accurately under a particular set of constraints.” The future rewards context, accountability, and domain knowledge more than raw conversion from one language to another.

Corporate writing becomes cheaper, but not automatically good

The video makes a related argument about corporate writing: companies may accept low-quality AI output because it is cheap, reducing the need for writers. There is a real economic mechanism here. If an organization only wants a passable internal memo, product description, or routine marketing variation, a model can create a first draft in seconds. A manager may decide that “good enough” is sufficient.

The official outlook is more measured. BLS projects little or no change for writers and authors from 2025 to 2035, with roughly 11,900 openings each year, mostly because people leave or change occupations. That is not a growth story, but it is not proof that writing disappears either.

The work most exposed is repeatable production with weak differentiation. The work most protected is connected to original reporting, a distinctive voice, subject expertise, editorial judgment, and responsibility for whether a claim is true. A company can generate ten plausible paragraphs cheaply; it still needs someone to know which paragraph is misleading, which customer will reject it, and which promise creates legal or reputational trouble.

That may produce a harsher market for beginners. An editor who once trained junior writers by assigning routine drafts may now use AI for those drafts and hire fewer trainees. At the same time, the value of a strong writer can rise when every organization is surrounded by cheap, forgettable prose. The scarce skill becomes not typing words but choosing what deserves to be said and making it trustworthy.

Software engineering can grow while the entry path narrows

Software is the clearest example of a mixed future. Dylan John describes engineering as a rising tide overall but says entry-level workers remain vulnerable. Current employment projections support the first half: BLS projects 10% growth for software developers, quality-assurance analysts, and testers from 2025 to 2035, with about 106,100 openings each year.

Growth does not settle the second half. A field can add jobs while changing the skills required to obtain them. If AI tools allow a small experienced team to produce more, companies may create new products and hire more specialists. They may also expect each new hire to arrive already capable of reviewing generated code, testing edge cases, understanding security, and working inside a real system rather than completing isolated classroom exercises.

The Stanford Institute for Economic Policy Research describes coding-heavy employment growth as slower in some respects but still positive, and reports that software-developer postings have been growing faster than other occupations over the last year. That is a useful counterweight to the claim that programmers are simply being replaced. It is not evidence that the traditional junior pathway is safe.

The future software worker may need to demonstrate judgment earlier. Knowing syntax will still help, but it will be less differentiating when an assistant can produce syntax. Understanding requirements, debugging unreliable output, protecting data, testing behavior, and explaining tradeoffs to non-programmers will matter more. The first job may be harder to get, while the people who clear that bar may have more leverage.

AI can narrow performance gaps while widening opportunity gaps

One of the video’s more hopeful claims is that AI can democratize skill. If a novice receives useful guidance at the moment of difficulty, the novice may improve more than an expert who already knows how to perform the task. A controlled workplace study summarized by the National Bureau of Economic Research found nearly a 14% productivity increase for customer-support agents using AI guidance. The improvement was 35% for the lowest-skilled and least-experienced workers, while the most experienced workers saw zero or small negative effects.

That is strong evidence for a mechanism, not a universal law. AI assistance can raise the floor on a task. It can give a new worker a checklist, a draft, or a worked example. It can also let a capable employee handle more cases. But access to the tool is not the same as access to opportunity. If companies respond to higher productivity by hiring fewer people, the workers who would have learned on the job may never receive the assistance that could have helped them.

This is the central contradiction of the AI jobs future. The technology may make individuals more capable while making the labor market less willing to give them a first chance. A junior worker can become more productive and still lose the job competition if the employer needs only one junior worker instead of three.

The current evidence says “mixed,” not “nothing”

The video argues that AI has produced limited aggregate employment impact and that adopting companies may even increase headcount. Multiple sources point in that direction, with important qualifications. Anthropic’s labor-market analysis finds no systematic increase in unemployment for highly exposed workers since late 2022, while finding suggestive evidence that hiring of younger workers has slowed in exposed occupations. Stanford’s review reports that firms adopting enterprise AI saw employment grow by 10% over the following two years, driven by high-spending adopters.

These findings do not mean AI creates jobs automatically. A company can use productivity gains to expand output, lower prices, enter a new market, or reduce staff. The choice depends on demand, competition, financing, management, and regulation. “AI adoption” is not one business decision with one labor outcome.

The Anthropic Economic Index offers a useful description of the present: in its initial analysis, AI use leaned toward augmentation, 57%, rather than automation, 43%. Usage was concentrated in software and technical-writing tasks. That split should not be read as a promise that augmentation will win forever. It is a snapshot showing that many people currently use AI to work with them rather than to replace every human step.

The video’s historical examples—ATMs, power looms, and other technologies—are helpful but limited. History shows that productivity can reduce the labor needed for a task while expanding demand for a service or creating related occupations. It also shows that the transition can be painful, unequal, and slow. A worker who loses an old job does not automatically receive the new one in the same town, at the same wage, or after the same training.

What to watch for over the next few years

First, watch entry-level hiring rather than only total employment. If total jobs remain stable while organizations reduce junior openings, the ladder is narrowing even without a headline unemployment spike. Track postings that ask for several years of experience for work that used to train beginners.

Second, watch the task mix inside occupations. A title such as “writer,” “developer,” or “translator” can remain on payroll while routine production is automated. Ask what the worker actually does, what decisions remain human, and who carries responsibility when the generated result is wrong.

Third, watch whether training expands. The OECD’s evidence points toward training and consultation as practical protections. A rising tide is more plausible when employers invest in teaching workers how to use tools, verify outputs, and move into adjacent responsibilities.

Fourth, watch the gap between productivity and headcount. If output rises but staffing falls, workers may experience the gains as insecurity. If output rises and firms expand into new markets, the same technology may feel like a growth engine. Company announcements should be checked against payroll, hiring, and workload rather than accepted at face value.

Finally, watch for evidence about the MIT enrollment claim in the video. MIT’s Course 6 catalog confirms that Fundamentals of Programming exists, but it does not verify the reported drop from 598 students to 198. Until an enrollment record is available, that number belongs in the category of an unverified creator claim, not a factual pillar of the forecast.

The future is a contest over the transition

The most defensible reading of Dylan John’s video is not that AI will kill all jobs or that nothing important is happening. It is that the first effects may be concentrated in routine tasks and early-career opportunities, while the broader economy changes more slowly and unevenly.

The evidence supports caution. A famous forecast of mass displacement is still a forecast. Translation and writing remain real occupations even as their routine work faces pressure. Software employment can grow while junior hiring becomes more selective. AI assistance can help inexperienced workers improve, yet employers can still use that productivity to reduce the number of people they hire. Current aggregate data show limited disruption, but aggregate stability can hide sharp losses for particular groups.

What ordinary people will see in the future depends on how organizations use the tools. The best outcome is not a world where humans refuse automation. It is a world where productivity gains fund better services, new work, and serious training rather than simply removing the first rung of every career ladder. The warning from the video is therefore worth keeping: the question is not only whether there will be work to do, but whether people will still be allowed to learn by doing it.

Sources

1. AI jobs danger: Sleepwalking into a white-collar bloodbath — Axios, May 28, 2025
2. Evaluating the Impact of AI on the Labor Market: Current State of Affairs — The Budget Lab at Yale, 2025
3. The impact of AI on the workplace: Main findings from the OECD AI surveys — OECD, March 27, 2023
4. Interpreters and Translators: Occupational Outlook Handbook — U.S. Bureau of Labor Statistics, accessed October 26, 2026
5. Writers and Authors: Occupational Outlook Handbook — U.S. Bureau of Labor Statistics, accessed October 26, 2026
6. Software Developers, Quality Assurance Analysts, and Testers — U.S. Bureau of Labor Statistics, accessed October 26, 2026
7. What is really happening to jobs? Separating AI hype from reality — Stanford Institute for Economic Policy Research, July 2026
8. Measuring the Productivity Impact of Generative AI — National Bureau of Economic Research, June 1, 2023
9. Labor market impacts of AI: A new measure — Anthropic, March 5, 2026
10. The Anthropic Economic Index — Anthropic, March 2025
11. Electrical Engineering and Computer Science (Course 6) — MIT Bulletin, accessed October 26, 2026
12. MIT Just Discovered What Jobs AI Actually Killed — Dylan John, source video