BCG published research this month projecting that 10 to 15 percent of U.S. jobs could be eliminated by AI within five years. That number gets the headline. The more useful number follows it: 50 to 55 percent of jobs will be reshaped rather than eliminated. Most professionals will keep their jobs. Their jobs will change.
The distinction between reshaping and replacing sounds like it might be the optimistic spin on a bad story. It isn't. Understanding the difference changes what you should actually do.
What "Replaced" Looks Like
Replacement happens when AI can perform most of the tasks that define a role, at acceptable quality, at lower cost than a human. The result is fewer people in that role, not just different people doing different things.
The clearest current examples are roles defined by high-volume, repetitive tasks. Tier-1 customer support is the most documented case: AI now handles 70 to 80 percent of inquiries without human involvement at companies like Block, which attributed a significant portion of its 4,000-person reduction in early 2026 to automated support systems. Recruiting coordination is another: the scheduling, ATS management, and candidate communications that once required coordinators are now handled by tools like Paradox, and the companies using them are not advertising for the same headcount they had two years ago.
In these cases, companies are reducing the number of people in the role. The work still exists, but it now requires a fraction of the human hours.
Replacement is real. It is also concentrated. It happens in roles where the task composition is narrow and the tasks themselves are well-defined enough for AI to execute reliably without a human in the loop.
What "Reshaped" Looks Like
Reshaping is what happens to roles where AI can take over some tasks, but the core of the job is too varied, too judgment-dependent, or too relational to automate fully.
A financial analyst who once spent 40 percent of their time pulling data and building standard reports is now spending that time on something else, because the AI does it faster. The job did not disappear. The task composition changed, and now the analyst needs to be better at the 60 percent that remains: the interpretation, the client communication, the judgment calls that require domain expertise and an understanding of what the numbers actually mean for the business.
For most professionals, this is the actual story. Not replacement, but a forced shift in what the job requires. That shift is not painless. Skills that used to matter less now matter more. Work that used to fill time will not be replaced by identical work at a lower level; it will be replaced by an expectation of higher-value output from the same number of working hours.
MIT research published this week reinforces the pattern. Rather than a sudden employment cliff, the evidence shows a slower, more uneven reshaping of work, varying significantly by industry, employer, and role. There isn't one AI transition happening at the same pace everywhere. There are thousands of them, at different stages.
The Question That Actually Matters
"Will AI take my job?" is the wrong question for most people. The more useful question is: which specific tasks in my role are automatable now, and what will I be expected to do with the time that frees up?
The answer depends on your role and employer, not on broad category-level claims. A financial analyst at a firm that has deployed AI for data work faces a different situation than one at a firm still running manual processes. A teacher in a district that has implemented AI tutoring tools is in a different position than one in a district that hasn't. The category-level averages obscure what actually matters: your specific task mix and your employer's specific deployment timeline.
The BCG data offers one more useful frame: the 10 to 15 percent elimination is not randomly distributed across the workforce. It is concentrated in roles where AI can perform the full range of tasks. The 50 to 55 percent reshaping is what happens to everyone else, which is most people. That math suggests the question for most professionals is not survival but adaptation.
What To Do With This
Take the tasks you do in a typical week and ask, honestly, which ones an AI could do at acceptable quality right now. Not theoretically someday. Right now, with tools that exist and are in active deployment.
For most knowledge workers, that list will include some things: standard report generation, templated communications, initial research passes, scheduling and coordination. For very few knowledge workers, it will include everything.
The tasks left over after that exercise are the ones worth investing in. The judgment calls. The relationship work. The synthesis that requires real understanding of context and stakes. These tasks are not automating quickly, and they are increasingly what employers need from the people they keep.
The reshape is already underway. Getting ahead of it means understanding your own task composition clearly enough to know which direction you're moving.