AI Is Not Coming in Waves, It Is Already Rising

What the Shape of AI Progress Reveals About the Future of Your Job 

Who This Is For: This lesson targets professionals and students who make decisions -- or advise organizations that do -- about how artificial intelligence will affect work. The primary audience includes business analysts forecasting operational efficiency, human resources professionals designing reskilling programs, workforce strategists evaluating labor risks and AI product managers who need to translate capability claims into deployment timelines. The secondary audience includes policy researchers studying structural unemployment, instructional designers building AI literacy curricula and upper-division undergraduate or graduate students in business, economics or technology programs. These individuals share a common problem: they encounter confident, often oversimplified narratives about AI displacing jobs overnight, yet they lack the empirical vocabulary to interrogate those claims. This lesson provides that vocabulary, grounded in a peer-reviewed study evaluating more than 17,000 real worker assessments of AI output across more than 3,000 labor-market tasks.

Real-World Applications

Industry Use Case: Workforce Planning at Scale

A multinational professional services firm uses task-duration profiling to identify which billable activities across its legal, financial advisory and consulting divisions fall within the rising-tide automation range. The firm maps each practice area against the study’s success-rate curves: tasks under four hours and highly text-intensive are flagged for near-term workflow redesign, while tasks requiring multi-week synthesis and domain judgment are classified as medium-term monitoring targets. This framework allows the firm to make headcount and reskilling decisions grounded in empirical capability data rather than vendor marketing claims or media speculation. The same approach is applicable to any organization whose work is predominantly knowledge-based and document-driven: from healthcare administration to government policy analysis to financial compliance.

Academic Connection

This use case operationalizes the study’s core empirical finding -- the flat logistic slope between AI success and log task duration -- by translating it into a screening tool for organizational decision-making. Researchers studying technology adoption can use the same framework to model sector-level exposure without relying on occupation-level automation probability scores that conflate task-level and worker-level effects.

The Problem and Its Relevance

The most dangerous story about AI and work is not that robots will take all jobs overnight, it is that because the disruption arrives gradually, organizations and workers can afford to wait before responding. A tide that rises slowly but continuously is not less consequential than a wave, it is simply harder to see until you are already standing in water up to your knees. The MIT FutureTech study examined over 3,000 labor-market tasks drawn from the U.S. Department of Labor’s O*NET classification, collecting more than 17,000 evaluations by workers with direct on-the-job experience, and found that AI improvement is broad-based and already substantial, not concentrated in a narrow set of dramatic breakthroughs. The second and equally unsettling finding is that current AI models already succeed at a significant share of real-world, text-based work tasks without any human edits, and this share is expanding rapidly across industries that have not yet reorganized themselves around that reality. Across all models in the study, AI achieved a roughly 60 percent success rate at producing output that a manager would accept as minimally sufficient without revision: a threshold that is far higher than most organizations have internalized in their hiring, training or technology investment decisions.

Why Does This Matter?

Understanding the shape of AI automation matters for the following reasons:

Flat improvement curves mean widespread exposure: Unlike a steep crashing-wave pattern that concentrates disruption in specific task categories, the rising-tide pattern documented in this study means that AI capability gains affect short tasks, medium tasks and long tasks simultaneously. No occupational domain is insulated.

Frontier AI models already outperform on multi-hour tasks: By the second quarter of 2024, frontier models achieved a 50 percent success rate on tasks that take humans approximately three to four hours to complete. By the third quarter of 2025, that threshold had extended to one-week tasks.

Newer models improve uniformly, not selectively: The study found that model releases after 2025 produced an approximately parallel upward shift across all task durations, meaning that the next generation of AI will not simply be better at short, simple work: it will be better at everything simultaneously.

Success rates at 80–95 percent are projected for most text tasks by 2029: Extrapolating current trends, the researchers estimated that LLMs will complete the majority of text-based labor-market tasks at a minimally sufficient quality level with success rates between 80 and 95 percent by 2029 under the assumption that current rates of AI capability growth continue.

Near-perfect automation requires additional years: The logistic shape of the success -- duration curve means that closing the remaining gap toward 100 percent success -- particularly for tasks requiring zero tolerance for errors will take considerably longer than reaching 80 percent, providing a measurable adjustment window.

Task-level automation does not equal job-level displacement: The study explicitly notes that automating individual tasks does not automatically eliminate positions, because occupational impacts depend on how tasks are bundled, how organizations restructure and whether automation raises or lowers the value of remaining human judgment within those roles. 

Three Critical Questions to Ask Yourself

• Can I distinguish between the crashing -- wave model of AI disruption -- where change is sudden and concentrated -- and the rising-tide model, where progress is gradual, broad and already underway across most text-based tasks?

• Do I understand why a 3.8-month doubling time for feasible task duration has fundamentally different implications for workforce planning than a single dramatic technological event would?

• Am I able to identify which specific roles, task bundles or professional domains in my organization carry the highest exposure to rising-tide AI automation, using task duration and text-intensiveness as screening criteria? 

Roadmap

Familiarize yourself with two core concepts: (1) the crashing-wave model, in which AI capabilities surge abruptly over a narrow set of tasks and (2) the rising-tide model, in which AI capabilities improve broadly and continuously across the full task-duration distribution. Then proceed through the group activity below. 

Group Activity (20 minutes)

Working in groups of three to four, complete all five steps below. Each step builds on the previous one. 

(i) Select an Occupational Domain

Choose one occupational domain from the O*NET categories referenced in the study (for example: Management, Computer and Mathematical, Legal, Healthcare Practitioners, or Business and Financial Operations). Identify three to five tasks within that domain that are predominantly text-based. For each task, estimate the human time required to complete it (five minutes, one hour, four hours, or one week).

Guidance: Use the task duration distribution from the study — most surveyed tasks fell between 20 minutes and 10 hours — as a calibration reference.

(ii) Estimate Current AI Success Rates

Using the study’s baseline finding that frontier models achieved approximately 60 percent success on tasks at the sample mean, apply the logistic slope of −0.31 per tenfold increase in task duration to estimate the predicted success rate for each of your selected tasks. A task taking one hour, for instance, would have a different predicted success rate than a task taking one week. Document your estimates and the reasoning behind them.

(iii) Map the Rising Tide to Your Domain

Identify which tasks in your selected domain would be affected first as success rates rise toward 80 percent by 2029. Discuss as a group: which tasks are currently at a 50 percent success rate, and therefore on the steepest part of the improvement curve? Which tasks are already above 70 percent, and therefore likely to reach near-automation levels soonest? Which tasks carry low tolerance for error, and therefore remain protected even as average success rates rise?

(iv) Design an Organizational Response

Using your analysis from steps (i) through (iii), propose a concrete organizational response. Your response must address: (a) which roles or task categories warrant immediate reskilling investment, (b) which task categories should be monitored quarterly rather than acted on immediately and (c) what human oversight mechanisms are necessary for tasks where AI success rates are high but error tolerance is low. Be specific about timelines and metrics.

Guidance: Avoid generic answers. The study explicitly warns that high success rates do not automatically translate into deployment -- last-mile integration costs, regulatory constraints, and economic attractiveness all shape adoption timelines.

(v) Challenge the Extrapolation

The study authors explicitly state that their 2029 projections assume continued AI capability growth at rates observed between 2024 and 2025, and that this represents an upper-bound scenario. As a group, identify at least two reasons why actual progress might slow, and explain how a slower pace would change your organizational response from step (iv). Consider factors such as compute scaling limits, algorithmic innovation slowdowns, and hardware constraints -- all cited in the study itself. 

Individual Reflection

After completing the group activity, write a brief individual response (three to five sentences per question) addressing the following:

• How did the rising-tide framing change the way you think about which roles in your field are at risk, and over what timeframe?

• The study found that larger models improve short tasks more than long ones, while newer models improve all tasks equally. What does this distinction imply for how organizations should think about the relationship between model size and deployment value?

• If AI success rates for tasks in your domain reach 80 percent by 2029, what specifically will still require human judgment at that threshold, and what institutional structures must exist to protect the quality of that judgment?

• The study notes that surveyed occupations slightly under-represent high-wage, high-education roles. How should this sampling limitation affect confidence in the findings for your specific professional context? 

The Bottom Line

The most consequential insight from this study is not that AI will eventually transform work, it is that the transformation is already measurable, already broad-based and already operating at task durations that encompass a significant share of what knowledge workers are paid to do every day. An organization that frames AI risk as a future event requiring future preparation is not being prudent, it is misreading the timeline in a direction that will make adjustment more disruptive, not less. The second insight is structural rather than predictive: the flat rising-tide curve means that no occupational domain receives a temporary exemption while adjacent domains absorb the impact. Legal, management, healthcare and scientific roles all exhibit meaningful AI success rates on text-based tasks, and all are improving in parallel. The question facing every professional is not whether their work will be affected but whether their organization has a framework for distinguishing which tasks benefit from AI acceleration, which tasks require human oversight regardless of AI performance, and which metrics will signal when a task has crossed the threshold from human-primary to AI-assisted. You have developed the AI literacy required to engage with this question substantively when you can articulate the difference between crashing-wave and rising-tide automation, apply task-duration profiling to real occupational data, interpret logistic success curves without conflating task automation with job displacement and make defensible workforce planning recommendations that account for both the pace and the structural shape of AI capability growth. This understanding is foundational for anyone whose professional role involves making decisions about talent, technology investment, organizational design or the future of knowledge work. 

#RisingTideAI   #AIWorkforceStrategy   #AIAutomationLiteracy   #FutureOfKnowledgeWork   #AICapabilityTracking