MIT report: AI will reshape text-based work gradually rather than eliminate professions all at once

A new MIT study offers a less apocalyptic view of AI's impact on the labor market. According to the team, changes are already moving quickly, but they resemble not a sudden collapse of professions, but rather a gradual rising of water: more and more text-based tasks become automatable, while companies and employees have time to adapt.

How It Was Measured

MIT researchers tried to answer not an abstract question—"when will AI replace people?"—but a more practical one: which work tasks can models already handle in real conditions. To do this, they took over 3,000 text-based tasks from the O*NET classification, which describes work in the American labor market, and gathered over 17,000 assessments from workers themselves. This is an important shift compared to laboratory benchmarks: the focus is not on tests for their own sake, but on whether a model can be trusted with a specific piece of work.

MIT separately compares two development scenarios. The first—"crashing waves"—is when AI suddenly becomes very strong in a narrow set of tasks and quickly disrupts entire categories of work. The second—"rising tides"—is when quality grows simultaneously across a broad front, without sharp jumps in individual niches. Based on the study results, the team found far more signs of the second scenario: progress is moving quickly, but more evenly and predictably.

What the Numbers Show

The key conclusion is that AI is already markedly useful for a significant portion of text-based work, if assessed by the threshold of "minimally sufficient." Researchers focused on 63% of work tasks in the U.S. economy that are text-based in nature and therefore, in principle, suitable for LLMs. Within this group, models were able to perform roughly 60% of tasks without human involvement at a level a manager would call acceptable. But if the bar is raised to truly strong results, the picture changes: only 26% of tasks were performed at an "excellent" level.

• In Q2 2024, AI completed tasks lasting roughly 3–4 hours with roughly 50% success
• By Q3 2025, this figure had grown to roughly 65%
• If the pace continues, by 2029 most text-based tasks will reach 80–95% success
• Near-flawless quality, according to MIT, will require waiting several more years

These figures explain well why the news looks both concerning and encouraging at the same time. On one hand, automation is moving faster than many would like. On the other—"minimally sufficient" does not equal "reliable," "high-quality," or "better than humans." In other words, business can already save time on drafts, document analysis, correspondence, and routine preparation of materials, but it's too soon to remove humans from the process entirely, especially in processes where errors are costly.

Why It's Not a Collapse

The authors directly challenge the popular scenario in which professions seemingly disappear in blocks after each model leap. Their argument is simpler: if AI's capability grows smoothly across many tasks at once, the market usually has time to respond. Companies don't lay off people on the day a new model launches—they still need to rewrite processes, adjust quality controls, distribute responsibility, and figure out where automation is even worth it. That's why technological progress and economic effect don't happen simultaneously.

"This is not a safeguard for workers, but AI progress can be noticed ahead of time," —

Neil Thompson, MIT.

This is the main good news from the report. It doesn't say that jobs are protected. It says that people, managers, and regulators have a chance to notice changes ahead of time, rather than wake up one day in a new reality. For office teams, this means not waiting for "day X," but gradually restructuring the human role: less mechanical text, more verification, task setting, decision-making, and responsibility for outcomes.

What It Means

For business, the conclusion is simple: it's time to stop debating whether AI will replace employees entirely and start breaking down work into individual tasks. That's where the main shift is happening. For specialists, the signal is equally clear: the winners will not be those who ignore AI, but those who integrate it into their process faster and retain what models still do inconsistently—context, professional judgment, and final quality.