Automation with Artificial Intelligence is one of the topics that generates the most anxiety in the working world these days.
Everyone has heard the story that AI is going to show up out of nowhere and wipe jobs off the map before people even have time to react.
It almost sounds like a disaster movie script, right? 🌊
But a recent study from MIT FutureTech came through with hard data that challenges that scenario, and the results are a lot more reassuring than most headlines out there would have you believe.
MIT FutureTech is an interdisciplinary research group formed by MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) in partnership with the MIT Initiative on the Digital Economy. And what they found wasn’t a wave that knocks everything down at once, but something more like a slowly rising tide, step by step, giving people time to notice the shift and reposition themselves.
And that difference changes how we can think about the future of work in a pretty significant way. 👀
How the researchers tested Artificial Intelligence
To reach these conclusions, the researchers did something pretty rigorous: they analyzed more than 60,000 evaluations made by real workers on AI-generated responses across more than 6,000 text-based work tasks. The central question driving the entire study was simple and powerful: does AI capability improve in sudden leaps or in a more gradual and continuous way?
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Unlike earlier studies that relied on coding tests or more deterministic benchmarks, this research went after tasks that are representative of the real world. The authors identified those 6,000-plus text-based tasks from the O*NET database maintained by the U.S. Department of Labor, and asked expert workers to evaluate the responses produced by AI. Each evaluation classified whether the AI output was minimally sufficient, average, or better than average, all without any subsequent human editing.
As Matthias Mertens himself, research scientist at MIT FutureTech and lead author of the study, put it in a pretty vivid way: imagine you are standing on a beach. In a crashing wave scenario, a few people get knocked down suddenly. But what they actually observed was the water rising around everyone, step by step. 🌊
AI capability is high and getting better
The study, fittingly titled Crashing Waves vs. Rising Tides, was authored by nine researchers from the MIT FutureTech lab, which is led by principal research scientist Neil Thompson. And the numbers they found are pretty revealing about where AI capability actually stands today.
Across all models analyzed, AI was already able to deliver minimally sufficient work on roughly 50% to 75% of text-based tasks, without needing any editing. And here is an interesting data point: when tasks became ten times longer, the AI success rate dropped by only about six to seven percentage points. That indicates the progress is broad and distributed, not concentrated in a narrow handful of specific tasks.
The pace of this evolution has been fast, too. Between the second quarter of 2024 and the third quarter of 2025, frontier AI models jumped from a 60% success rate on tasks that take humans about 1.5 hours to complete, to a success rate above 70%. Overall, the researchers calculated that the rate at which AI fails to complete a task drops by half every 2.2 to 2.8 years. That is consistent, measurable progress. 📊
The difference between task automation and job automation
One of the most important points the study brings up is the distinction between automating a task and automating an entire job. It might sound like a minor detail, but it changes everything when it comes to understanding what is actually happening. Most jobs are made up of a mix of tasks, some more routine and others that require judgment, creativity, interpersonal skills, and constant adaptation to context.
As Mertens was careful to point out, the success rates should not be interpreted as though a corresponding share of tasks could or should be automated today. In the study, the AI was given all the information needed to complete each task. But in real life, that information is often incomplete, difficult to integrate, or subject to regulatory constraints, which significantly complicates the feasibility of adopting AI in actual work environments.
Think, for example, about a financial analyst who spends a big chunk of the day consolidating spreadsheets, cross-referencing data, and generating standardized reports. Those specific tasks are exactly the kind of thing Artificial Intelligence tools can already execute efficiently. But the part of the job that involves interpreting the numbers within the context of the business, talking to clients, understanding market nuances, and recommending strategic paths is still far from being handed off to a machine without supervision. So in practice, what changes is how that professional distributes their time and energy, not necessarily whether their role exists.
This understanding is crucial because it helps direct the efforts of anyone who wants to position themselves well in the face of these changes. Instead of trying to flee from areas that seem generically threatened, it makes a lot more sense to map out which specific parts of your daily work are most susceptible to automation and which are the ones where your presence, judgment, and relational skills make a real difference. And that is exactly where the MIT research becomes a valuable resource, because it offers a framework for thinking about this in a structured, evidence-based way. 📊
The impact of automation varies a lot by industry
Another important finding from the study is that automation does not affect all occupations in the same way. This reinforces the idea that there is no one-size-fits-all formula when it comes to adapting to the advancement of AI. The researchers found a wide variation in success rates depending on the type of work analyzed.
To give a concrete example, AI success rates were below 50% in text-based legal tasks, while they exceeded 70% in tasks associated with installation, maintenance, and repair. This suggests that AI capability is advancing at different speeds depending on the occupation, which means, in practical terms, that the window of time for adaptation will be wider in some fields than in others. Those working in areas with higher exposure have more reason to start moving sooner, while others will have a bit more breathing room. 🤖
Preparing for change: what this means in practice
Talking about preparing for change in the context of Artificial Intelligence can sound vague, but there are very concrete paths that make sense for most people. The researchers themselves estimated that most text-based tasks could reach success rates between 88% and 97% by 2030, at a minimally sufficient quality level. Near-perfect performance, on the other hand, is still likely several years beyond that.
As Mertens put it, the key takeaway is that this gives us time. Precisely because the change is gradual, there is a real window to adapt. And making the most of that window involves developing what experts call AI literacy, which basically means understanding enough about how these tools work to be able to work alongside them more effectively, without needing to become a programmer. That includes knowing how to write good prompts, understanding the limits and failure modes of these systems, recognizing when a tool’s output needs human review, and incorporating these technologies into your workflow in a smart way.
Another important move is strengthening skills that, at least for now, are beyond the reach of AI capability. Complex communication, leadership, empathy applied to work, critical thinking in ambiguous scenarios, and creativity that goes beyond recombining existing patterns remain essentially human competencies. The good news is that these skills develop through experience, practice, and intentional reflection on your own work. 🧠
What AI capability still cannot do
Part of the doomsday narrative around Artificial Intelligence comes from overestimating what current systems actually deliver autonomously and reliably. The researchers themselves are careful to point out that their projections assume AI will continue improving at its recent pace, and that these numbers should be seen as a ceiling scenario. Near-perfect performance is still expected to take considerably longer, especially in contexts where errors carry serious consequences.
Language models are extremely good at generating coherent text, summarizing information, and answering questions within well-defined contexts, but they can fail in unexpected ways when the subject requires multi-step chained logical reasoning or precise factual verification. This phenomenon, known as model hallucination, is a real technical limitation that still demands constant human oversight in any critical application.
On top of that, AI-based automation struggles with what is known as tacit knowledge, the kind of know-how that an experienced professional carries but has never been explicitly documented. A veteran mechanic who identifies a problem by the sound of an engine, a doctor who senses something is wrong before the test results confirm it, a teacher who knows exactly how to approach a struggling student. That kind of practical intelligence, built over years, is very difficult to transfer to an AI system with the approaches available today.
As Mertens himself put it, by the end of the decade these models will be extremely capable across many tasks, but that does not mean entire occupations are simply going to disappear. Many jobs include a set of tasks that is beyond the reach of language models, so the most likely outcome is that work will be reorganized between humans and AI. 💡
The real shift is to stop asking whether AI is going to take your job and start asking which parts of your work it can do better than you, and what that frees you up to do with more depth and impact.
The MIT FutureTech study did not come to say that everything is going to stay exactly the way it is. Automation with Artificial Intelligence will keep advancing, it will transform roles, it will demand adaptation, and it will create new opportunities we cannot even fully imagine yet. But it came with data showing that this process has a much more manageable pace than the collective imagination tends to paint, and that the window of time to reposition yourself is real, concrete, and wider than the end-of-the-world headlines suggest. That is not a reason to completely relax, but it is definitely a solid argument for swapping out paralyzing anxiety for an active curiosity about what is coming next. 🚀
