Corporate leaders spent the past two years promising that generative AI would turbocharge output and slash costs. Instead, some of the most closely watched analysts now argue that the technology is barely moving the productivity needle and, in some cases, is actively getting in the way of real work. The gap between the hype and the hard numbers is widening, and the people paid to track these trends are starting to say so out loud.
At the center of that backlash is a top analyst warning that AI is, in their words, completely failing to deliver the productivity boom investors were sold. I see that claim echoed in economic forecasts, workplace studies, and even internal enterprise post‑mortems that describe a wave of “Good Ol’ Procrastinating” dressed up as digital transformation rather than genuine efficiency gains.
The analyst who says AI is “Completely Failing”
The sharpest critique comes from a veteran technology analyst who argues that the current wave of enterprise tools is Completely Failing to deliver the promised Boost Productivity. In their view, executives have been sold a story in which chatbots and copilots quietly handle drudge work while humans focus on strategy, but the lived reality inside many firms is closer to Good Ol’ Procrastinating with extra steps. Workers are spending time experimenting with tools that still hallucinate, misread context, or require so much oversight that any time saved on drafting is lost in verification.
That skepticism is reinforced by a separate warning that the very introduction of AI into workflows can slow people down rather than speed them up. Citing a widely discussed MIT experiment, the analyst notes research suggesting that “95 percent” of workers in some settings saw no meaningful productivity improvement once AI tools were added to their tasks. When the overwhelming majority of users fall into that category, it is hard to argue that the technology is, at this stage, a broad-based engine of efficiency rather than a niche accelerator for a small slice of work.
Agents, automation, and the Center for AI Safety reality check
Behind the scenes, companies have been betting that fully autonomous agents will do what chatbots have not: quietly take over entire workflows. Yet when researchers at the Center for AI tested AIs designed to automate entire tasks, the results were described as “not looking too hot.” These systems struggled with multi‑step instructions, failed on edge cases, and often required human babysitting that erased any theoretical labor savings. For organizations that had quietly penciled in automation of up to some 10.4 million roles, that is a sobering signal that the technology is not yet ready to shoulder the load.
Even the more optimistic commentary around AI’s economic impact tends to acknowledge that the payoff will be slow and uneven. A research note on The AI megatrend argues that the first half of the year may be softer for productivity, in part because companies are still digesting stagflationary shocks and tariff changes that complicate capital spending. Analysts there say AI-driven capital investment is reshaping stock markets, but they stop short of claiming that those investments have already translated into measurable output gains on factory floors or in service jobs.
Macroeconomic forecasts without an AI miracle
At the macro level, the story is similar: growth is improving, but not in a way that screams “AI revolution.” Goldman Sachs Research projects that the United States will outperform many economist forecasts in 2026, with a baseline view that a better policy environment will be enough to boost hiring and keep growth in the 3% to 3.25% range over the cycle. That is solid, but it is not the kind of step‑change one would expect if AI were already unlocking a once‑in‑a‑generation productivity surge across the economy.
Another note on What 2026 holds for tech stocks notes that analysts see AI-driven capital investment reshaping market leadership, with some sectors pulling ahead as they pour money into automation. Yet even there, the language is about positioning and valuation rather than realized efficiency. Investors are betting that AI will eventually justify its price tags, but the hard data that would prove that thesis is still thin.
Inside the enterprise “honeymoon phase” hangover
Inside companies, the mood has shifted from exuberance to something closer to buyer’s remorse. A widely shared analysis of enterprise deployments argues that the “honeymoon phase” of corporate AI is over, with leaders realizing that the real challenge is not model implementation, it is workforce transformation. One summary of that shift, framed around Goldman Sachs and its view that Integration will Redefine Workforce dynamics in 2026, stresses that the real bottleneck is change management, not GPU capacity.
That disconnect shows up in the day‑to‑day experience of knowledge workers. A report on AI paradoxes notes that, For the knowledge sector, the outlook is nuanced, and that While workers are being encouraged to embrace AI, a recent MIT study found that much AI-assisted work requires unnecessary extra work. Instead of freeing people to focus on higher‑value tasks, the tools often generate drafts, summaries, or slide decks that then need to be painstakingly checked for errors, bias, or simple irrelevance.
“Workslop” and the cost of low‑quality output
Researchers have started to give that phenomenon a name: workslop. Writing in Harvard Business Review, the researchers define workslop as “AI-generated content that looks polished but doesn’t actually meaningfully advance a given task.” In practice, that might be a sales email that sounds slick but misstates the product, or a project plan that is formatted perfectly but omits key constraints. Teams then spend hours fixing or rewriting this material, often concluding that they would have been faster if they had done the job themselves from the start.
One detailed study, highlighted by Stanford University researchers and a workplace performance firm, reports that workers say they spend a significant share of their day fixing low‑quality AI‑generated “work.” Another investigation into AI-generated “workslop” finds that the end result inside teams is “confusion, annoyance, wasted effort and then some serious layers of judgment,” with one estimate putting the cost of this dynamic at $9 million in wasted time per year for a single large organization.
More from Morning Overview