How AI Is Transforming Workflows and Redefining Professions
Artificial Intelligence has already become part of the daily routine at many companies, but most are still using the technology in a pretty limited way. Writing emails, summarizing documents, generating code snippets… these are one-off, isolated tasks, and as helpful as they are, they’re far from showing what AI can really do.
What if the problem isn’t the tool itself, but the way companies are thinking about how to use it?
That’s exactly what new research from MIT Sloan School of Management puts up for discussion. The study, titled Chaining Tasks, Redefining Work: A Theory of AI Automation, brings a different perspective on how AI creates value, and the answer isn’t in individual tasks. It’s in workflows as a whole.
The paper was written by Peyman Shahidi, a doctoral candidate at MIT Sloan, in partnership with professors Mert Demirer and John Horton, also from MIT Sloan, along with Nicole Immorlica from Yale University and Microsoft, and Brendan Lucier, a senior researcher at Microsoft Research. The team brings together experts in labor economics, theoretical computer science, and market design, which helps explain why the research goes well beyond the obvious takes on automation.
The researchers argue that the real impact of Artificial Intelligence happens when you look at the full sequence of tasks within a role, not at each step separately. This changes the conversation about automation, efficiency, and even the future of professions in a big way. Let’s break down how this research could transform the way companies think about AI 👇
The Wrong Way to Use AI in Business
Most organizations still treat Artificial Intelligence as a collection of separate tools, each solving a specific problem here and there. An assistant for answering emails, another for generating reports, another for suggesting code snippets. In practice, this creates a fragmented experience where AI steps in for a moment and then exits the picture while human work takes back control.
This model works, but it works poorly. When you plug AI into isolated points of a process, the efficiency gain is real but limited. It’s like swapping out a single part in an engine without looking at the whole system. The car might run a tiny bit better, but you never truly tap into the engine’s full potential.
And that’s exactly the central point the MIT Sloan study raises: the wrong unit of analysis is distorting the way companies measure and apply the value of AI. As Peyman Shahidi put it pretty directly, the central question is no longer simply how AI improves an individual task. The goal now is to understand AI’s effect at a broader level, at the system level, rather than as one-off productivity tools applied task by task.
When the focus is on individual tasks, decisions about where to apply automation also get distorted. Companies end up prioritizing what’s easiest to automate instead of what would generate the most impact if automated within a larger context. The result is technology usage that looks impressive in presentations but barely changes the business’s day-to-day operational reality.
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.
Workflows Are the Key, Not Tasks
The big shift the study proposes is simple to understand but profound in its implications: instead of asking which tasks can AI handle, companies should be asking which complete workflows can AI transform.
This change in perspective seems small, but it changes everything. A workflow is a sequence of connected, interdependent tasks that together produce a result. The research built models of how these tasks are sequenced and connected in real-world workflows, establishing a new framework for understanding how work actually happens. When you automate just one of those tasks, the workflow still requires human intervention at other stages, and that’s where the real gains shrink significantly.
The researchers argue that it’s in the connection between tasks, in what they call task chaining, that AI shows its maximum value. Imagine a data analysis process that involves collection, cleaning, interpretation, and report generation. If AI only handles collection, someone still needs to clean the data, interpret the results, and put the report together. But if it handles the entire chain, the efficiency gain is exponential, not linear. This transforms AI from a support tool into a central agent within organizational processes.
Not all chaining works the same way
An important detail from the research is that not all task chains are equal. When adjacent tasks are well-suited for AI, they can be effectively grouped into a continuous flow. However, when a single step in the sequence is too difficult for AI, it can break the entire chain.
Shahidi was pretty direct on this point: if one of the tasks is extremely difficult for AI, that single task will compromise the whole operation. This means that when planning automation, it’s not enough to look at the easy tasks. You need to identify the bottlenecks that could interrupt the automated flow and figure out how to deal with them.
This finding leads to a new work design principle that the research highlights with considerable emphasis: how tasks are grouped matters just as much as which tasks are automated. It’s a paradigm shift that requires rethinking the structure of processes before rushing to implement any AI tool.
The example that makes it all click
The researchers use a really straightforward example to illustrate this idea: the difference between a lecturer who teaches large classes and a tutor who works one-on-one with students. Both roles involve similar tasks, like explaining concepts, creating exercises, and assessing comprehension. But the workflows are completely different.
The lecturer prepares content in advance, follows a more predictable structure, and can clearly separate stages. This makes it easier to automate parts of the process. The tutor, on the other hand, works in a continuous back-and-forth dynamic with the student, reacting in real time and adapting instruction to each response. This workflow is much harder to chain for AI.
As Shahidi explained, the ability to automate workflows using AI is very limited in that second role. It’s precisely the way tasks appear within each profession’s workflow that determines automation potential, not the tasks themselves.
This concept also redefines how we should think about which roles are most impacted by Artificial Intelligence. They’re not necessarily the roles with the easiest tasks to automate. They’re the roles whose workflows have the greatest chaining potential, meaning where tasks connect in a logical and sequential enough way for AI to operate end-to-end with little or no human interruption.
Why System Efficiency Beats Perfection at Each Task
One of the most counterintuitive findings from the research is this: AI doesn’t need to outperform humans at every individual task to generate significant value. In fact, organizations can benefit from assigning entire task chains to AI, even when humans would do some of those steps better in isolation.
The reason is coordination cost. Every time work passes from AI to a human, it needs to be reviewed, validated, and adjusted. These checkpoints slow down the system as a whole. On the other hand, allowing AI to handle a sequence from start to finish eliminates friction, reduces handoffs between people and machines, and speeds up delivery, even if the quality of each individual step is slightly lower than what a human would produce.
Shahidi pointed out that in this model you save human time, and that removing repeated oversight can offset marginal performance differences between human and machine at each specific task.
This completely changes how leaders should evaluate AI. Instead of measuring whether it excels at each individual step, the focus should be on whether it improves the efficiency of the complete workflow. This perspective also reinforces the importance of task adjacency. When AI-suitable tasks are grouped together, they can be executed in a single continuous flow. When they’re scattered or interrupted by tasks AI can’t handle well, the benefits drop considerably.
The Real Impact on Organizational Design
When you start looking at AI through the lens of complete workflows, the consequences for organizational design are huge. Companies need to rethink not just where the technology fits in, but how teams are structured, which roles make sense, and how work is distributed between humans and automated systems. This isn’t a futuristic discussion. It’s a present-day necessity for any company that wants to compete seriously in the coming years.
Historically, professional roles were defined as bundles of tasks that are most efficiently performed by a human. AI changes that equation by reducing the cost of certain activities and enabling new combinations of work. If AI can automate several routine tasks within a role, employees can take on additional responsibilities, often higher-value work that requires human judgment and critical thinking.
The study suggests that the organizations pulling ahead aren’t the ones with access to the most sophisticated AI tools, but the ones redesigning their internal processes to take advantage of task chaining. This requires an honest reading of how work actually happens inside the company, mapping every step, every dependency, every decision point. It’s an organizational design exercise before it’s a technology exercise, and many companies still haven’t realized that.
There’s also a direct consequence for the people inside these organizations. As workflows are automated more broadly, human roles shift. Instead of executing tasks, people move to supervising systems, making decisions at critical points, and handling exceptions that AI still can’t resolve. This doesn’t necessarily mean fewer jobs, but it does mean different jobs, with different skills and a completely new relationship with technology in the workplace. Over time, this could reshape how work is distributed across teams and entire functions within companies.
Patience Is Part of the Process
One point the research highlights that many companies overlook is the question of time. For business leaders, AI adoption stops being a purely technological decision and becomes a broader organizational design challenge. And this kind of change doesn’t happen overnight.
Many companies expect quick returns on their AI investments, but the research indicates that real gains often only show up after organizations adapt their workflows and build enough capacity to operate under this new model.
Shahidi was pretty clear about this: until you reach that maturity point, the costs of adopting AI outweigh the gains. Only after crossing that threshold does restructuring work around AI start delivering measurable benefits. In other words, there’s an investment and adaptation period that companies need to go through before reaping the real rewards.
What Actually Changes in Practice with This Perspective
Adopting this workflow-oriented perspective has very concrete practical implications. The first step is mapping the company’s processes with fresh eyes, identifying not where AI can help with a specific task, but where it can take over an entire sequence of activities with coherence and consistency. This mapping often reveals opportunities that go unnoticed when the focus is on isolated tasks, and many times these opportunities are sitting right in the most repetitive and structured processes in the business.
The second step is understanding that workflow-oriented automation requires integration between systems. There’s no point in having excellent AI for text generation if it can’t connect to the CRM system, the customer history, and the company’s business rules within the same process. The technology infrastructure needs to be ready to support this chaining, and that often requires investments in data integration and systems architecture that go beyond simply adopting a new tool.
The third step is rethinking the mindset around what it means to adopt AI. The quote that probably best captures the spirit of the research came from Shahidi himself: it’s not about how I’m going to introduce AI into my existing workflow, but about how I can redesign my workflow in a way that’s more AI-friendly. This flip in logic is what separates companies using AI superficially from those truly transforming their operations.
Finally, there’s a necessary cultural shift. Teams need to understand that AI’s role isn’t to replace what they do, but to take on the parts of the work that can be structured and automated so people can focus on what truly requires human judgment. Organizations that treat AI as a tool that fits into existing processes may see incremental improvements. But those that rethink the structure of work, grouping AI-compatible tasks, reducing unnecessary handoffs, and redesigning flows, have a much better shot at unlocking this technology’s full potential.
When this mindset is firmly in place, technology adoption happens with far less resistance and far better results. And that’s when the real potential of Artificial Intelligence truly starts showing up in the business numbers. 🚀
