AI adoption rates look weak — but the data hides a bigger story

Is AI in a bubble?

That’s the basic yet seismic question on a lot of people’s minds. But here’s the thing: It’s oversimplified, attempting to color an unprecedentedly gray moment either black or white. And what does the query even mean? If you’re asking about whether or not the valuations of certain AI startups and the companies that supply them are overvalued relative to their current financials, there’s a strong case for answering in the affirmative. If you’re asking whether the hype over AI has raced ahead of the technological landscape in regards to it attaining artificial general intelligence or rapidly destroying the labor market — the answer might be “probably.”

But if you’re asking whether AI will ultimately fizzle out and go down in history as the fever dream of a science-fiction-obsessed Silicon Valley cult, willed into existence by billions of FOMO-fueled venture capital dollars, the answer is undeniably “no.”

Where is AI adoption, anyway?

Perhaps the best data available to help us answer any form of the omnipresent “bubble” question is AI adoption rates. Are individuals and businesses actually using AI? Are they deriving benefits? Are they integrating it into what they do each day? 

Let’s look at businesses first. QuantumBlack, McKinsey’s AI arm, published a report in June showing that 8 in 10 companies use generative AI — so the technology seems to be catching on pretty quickly.

However, an MIT NANDA report made public two months later appeared to reaffirm skeptics’ canary calls. The researchers analyzed 300 publicly disclosed AI initiatives at various companies, interviewed representatives from 52 organizations, and surveyed 153 senior leaders at major industry conferences.

“Despite $30-40 billion in enterprise investment into GenAI, this report uncovers a surprising result in that 95% of organizations are getting zero return,” the authors reported.

That headline-friendly finding spread quickly, trumpeted by a range of media sources as evidence that the presumed AI bubble is about to burst. The previously published QuantumBlack report also indicated that AI had a similarly lackluster effect on companies’ bottom lines, though it failed to generate buzz at the time.

Moreover, starting in June, companies of all sizes reduced their use of AI, according to the U.S. Census Bureau.

Both the MIT researchers and QuantumBlack analysts would say that this is a temporary blip, however. Given how the media widely covered their reports, one might think they hold negative views of AI’s practicality in the business world, but the opposite is true. In reality, both teams gushed that AI is already transforming the way people work and is set to overhaul how businesses operate. 

“For the first time in human history, you can manipulate technology with human language, not a computer science language,” Alexander Sukharevsky, a Senior Partner at McKinsey and the global leader of QuantumBlack, told Big Think in an interview. “It’s always hard to predict the future,” he cautioned, adding: “If I look at the current adoption and willingness of organizations to go for a transformation, it’s the highest ever in my career.”

Two trends help fuel Sukharevsky’s and the MIT team’s optimism over AI. The first is that while most enterprises haven’t derived tangible advantages from adopting AI just yet, individuals have.

An AI “shadow economy” at the workplace

“As of late 2024, nearly 40 percent of the U.S. population aged 18-64 uses generative AI,” a trio of economists reported earlier this year. “Twenty-three percent of employed respondents had used generative AI for work at least once in the previous week, and 9 percent used it every work day.”

If those statistics don’t seem groundbreaking at face value, consider them in a historical context.

“Relative to each technology’s first mass-market product launch, work adoption of generative AI has been as fast as the personal computer (PC), and overall adoption has been faster than either PCs or the internet,” the authors noted. Granted, PCs and the internet were harder to initially embrace due to cost and difficulty of setup, but AI adoption is right in line with two pivotal products of our current technological era, evincing its long-term transformative potential.

The MIT team witnessed this firsthand while researching their report. Though AI initiatives floundered at the business level, AI itself was widely used among the workforce.

“AI is already transforming work, just not through official channels. Our research uncovered a thriving ‘shadow AI economy’ where employees use personal ChatGPT accounts, Claude subscriptions, and other consumer tools to automate significant portions of their jobs, often without IT knowledge or approval.

The scale is remarkable. While only 40% of companies say they purchased an official LLM subscription, workers from over 90% of the companies we surveyed reported regular use of personal AI tools for work tasks. In fact, almost every single person used an LLM in some form for their work.”

These employees had a solid grasp of what it would take for AI to succeed at a grander, enterprise-wide scale: learning and memory. AI must be able to retain information over extended periods of time and adapt to changing circumstances. Which brings us to the second reason that both the MIT and QuantumBlack teams are certain of AI’s integral future in business despite its tepid success so far: AI systems with those abilities are here and are now starting to be rolled out.

Unleash the agents

The overwhelming majority of AI systems used in businesses today are easy-to-implement, shareholder-friendly chatbots. But while these might slightly improve productivity at the worker level — summarizing meetings, making images, or writing emails — they aren’t going to deliver the revolutionary change to workflows that drives major returns on investment.

Chatbots are primarily conversational tools designed to answer questions from a predefined knowledge base. AI agents, on the other hand, are sophisticated, autonomous systems that can analyze information, make decisions, and take actions to achieve goals. Fundamentally, chatbots are reactive, whereas AI agents are proactive. A chatbot is like a calculator — essentially a tool. An agent is a collaborator with a calculator of its own.

“Think about an infinite army of interns that is able to do a lot of simple tasks,” Sukharevsky said of agentic AI. “But you need to orchestrate them, you need to upskill them, you need to provide them with the right information.

Agents are in the early stages of implementation, and these have the potential to deliver on the grander promise of AI.

“Don’t expect the first time to see a masterpiece,” Sukharevsky said. “But it actually evolves. What we see today is the worst … If you look at the technology a year ago, two years ago, and if you look at the technology today, or twelve months down the road, you get completely different outcomes.”

Agentic AI’s rapid progression evinces its shift toward widespread, practical use in enterprises. The MIT researchers found more evidence for this trend.

“In the next few quarters, several enterprises will lock in vendor relationships that will be nearly impossible to unwind,” they wrote.

The consulting firm Source echoed what the researchers were seeing in a trend report published in mid-September. Some 55% of its clients planned to invest in organizational restructuring during the next 18 months, almost entirely due to AI.

“It is impossible to hide from the impact of AI,” the authors wrote. “Few organisations—if any—do not have a roadmap for AI implementation.”

What might agentic AI look like when integrated into a business? QuantumBlack analysts provided a few examples. In e-commerce, agents could observe a user’s behavior, shopping cart content, and purchase history to offer product recommendations in real time. In supply chain management, agents synced to internal and external data sources could continuously forecast demand, allocate stock, determine optimal transport, and more. At a retail bank, agents could formulate credit-risk memos on prospective clients, a task that can currently take a human worker two to four days to complete. Instead, that person can check over the AI’s work and then make the loan decision.

What about us?

If AI agents are essentially digital workers, and you can create a ton of them, where does that leave human workers? 

The broader topic of job loss in the era of AI is widely covered, heavily debated, and nearly impossible to predict in the long term. The only thing we can say for sure is that the nature of work is going to change.

The MIT researchers did hear about some workforce effects directly from company leaders.

“Organizations that have crossed the GenAI Divide are beginning to see selective workforce impacts in customer support, software engineering, and administrative functions … Executives were hesitant to reveal the scope of layoffs due to AI but it was between 5-20% of customer support operations and administrative processing work in these companies.”

The researchers also asked workers in these companies for their opinions.

“Concerns about workforce impact were far less common than anticipated. Most users welcomed automation, especially for tedious, manual tasks, as long as data remained secure and outcomes were measurable.”

Sukharevsky briefly described the role of a human worker in an agentic AI future.

“Your role is to find the right team members, the right skills, the right data, and to put them against the right mission. If you invest enough time, that [AI] intern all of a sudden becomes an expert one day — your peer — and it can add a lot more value.”

Effects of the continuing AI rollout will not be felt equally in all sectors, the MIT team learned. In healthcare, energy, and advanced industry, executives didn’t anticipate any hiring reductions over the next five years. In technology and media, however — where AI is coding, producing images, writing scripts, and crafting video — staff reductions are underway now and more are expected in the near future.

But new jobs and industries will spring up thanks to agents, Sukharevsky said. He’s already seeing it.

“Newcomers are creating completely new business models that you couldn’t imagine in the past, and you couldn’t imagine them because the unit economics couldn’t fly, now they finally fly.”

Sukharevsky stressed that it will take years before agentic AI is incorporated into large, established businesses. Legacy software architecture, human perception, management, and governance are all barriers to successful adoption. 

“You’re seeing mixed results because you need to go ‘all-in’ to get the outcome. It’s difficult, it’s expensive, and it’s not immediate.”

But it is definitely coming. QuantumBlack offers some prebuilt agents and a builder software product for companies to create their own.

“What we’re trying to do is close the gap between the promise of AI and the reality,” Sukharevsky said. “We already have a few dozen agentic transformations under the belt and ongoing, so this is real.”

The answers we seek

Humans generally aren’t hardwired for patience. When faced with uncertain circumstances, we grasp for certainty. After the MIT team published their report, many onlookers grasped onto the notion that AI wasn’t yielding financial results; thus the anxiety-inducing structural change to work and life that AI represents might never materialize. But delving deeper into the report and reading others like it makes clear that AI’s broad effects are only beginning to manifest. 

Change is undeniably afoot. AI’s rise is often compared to the Industrial Revolution. Whether AI will actually result in comparable societal upheaval and economic reorganization is anybody’s guess. The only certainty is that we won’t get the answers we seek in a single report, from the latest adoption data, or even over a matter of months or years. The Industrial Revolution played out over eight decades. Even if the AI economic transformation plays out four times faster, we still have a long ride to go.