A new report highlighted by New Jersey Business Magazine argues that AI is already capable of unlocking about $4.5 trillion in U.S. labor productivity by taking on (or assisting with) a huge share of day-to-day work tasks. The report is Cognizant’s “New Work, New World 2026,” and it’s based on a re-check of roughly 18,000 tasks and 1,000 jobs from the O*NET labor database.

The headline number is attention-grabbing, but what I found more interesting is the idea of “exposure scores,” basically a measure of how much of a job can be assisted or automated by AI. The report says the average exposure across jobs is now 39%, which is far higher than the old projection for the early 2030s, and that the exposure score is rising much faster than previously expected. In plain English: this shift isn’t coming “someday,” it’s already here and spreading into roles people assumed were relatively safe.

Some of the examples in the article are pretty wild. Legal work is cited as jumping from 9% exposure to 63%. Education moves from 11% to 49%. Healthcare practitioners go from 10% to 39%. Even C-suite roles (including CEO) are listed as seeing big increases. Meanwhile, the report suggests manual-labor roles may be more affected than many expected, with transportation and construction exposure rising as well.

Here’s the part I agree with most: AI isn’t a “blanket solution.” The article emphasizes that human involvement, judgment, and adaptable operations are still required to capture real value. That matches what I see in practice. AI can be great at drafting, summarizing, classifying, searching, and generating options, but turning those outputs into decisions (and then into results) still depends on context, accountability, and good process design.

The report also notes a key constraint: AI can’t automate a big chunk of work in management, business/financial operations, and administrative tasks, and that “one-size-fits-all” approaches won’t get companies to the promised productivity gains. To me, this is where the conversation should shift. The “new feature” isn’t just smarter models. It’s the emerging operating model: teams that treat AI like a co-worker (with clear boundaries), invest in employee upskilling, and redesign workflows so AI actually removes friction instead of creating more review loops.

My take: the $4.5T number is plausible as “potential,” but the real story is how uneven the outcome will be. Companies that pair AI with training, governance, and workflow changes will compound productivity. The ones that bolt it on as a shiny tool will mostly get noise: more content, more drafts, more meetings, and not much more output.