There’s a strange phenomenon happening in workplaces right now, and it’s one that doesn’t quite make sense at first glance. The people seeing the biggest productivity gains from AI developers writing code 30% faster, marketers churning out campaigns in half the time, analysts completing complex research in minutes instead of days are also the ones staring at the ceiling at 3 AM, wondering if they’ve just automated themselves out of a job.
Welcome to what researchers are calling the AI productivity paradox of 2026. The better you get at using AI, the more worried you become about your future. It’s a twist nobody saw coming.
The Data That Doesn’t Add Up
Let me start with some numbers that paint a confusing picture.
According to a February 2026 study from the National Bureau of Economic Research that surveyed 6,000 executives across four countries, more than 80% of companies reported zero measurable productivity gains from AI. Not small gains. Zero.
Yet at the same time, U.S. labor productivity grew 2.7% in 2025 nearly double the 1.4% annual average of the previous decade. Goldman Sachs found that companies successfully measuring AI’s impact on specific tasks reported productivity improvements of around 30%. Software developers using tools like GitHub Copilot and customer service agents equipped with AI response suggestions both showed median gains of about 30%.
So which is it? Is AI revolutionizing productivity, or is it just expensive noise?
The answer, it turns out, is both. And that contradiction is exactly what’s keeping people up at night.
The Workers Who Actually See the Gains
Let’s talk about the people for whom AI isn’t theoretical it’s already transforming their daily work.
A recent survey from Anthropic found that workers in roles most exposed to AI displacement developers, IT professionals, data analysts, content creators are the same people reporting the most dramatic productivity improvements. They’re not just dabbling with ChatGPT to write better emails. They’re fundamentally restructuring how they work.
One software engineer put it bluntly in the survey: “Like anyone who has a white collar job these days, I’m 100% concerned, pretty much 24/7 concerned, about losing my job eventually to AI.”
Read that again. This person isn’t some Luddite afraid of technology. They’re using AI tools every single day. They’re seeing firsthand how well these systems work. And that’s precisely why they’re terrified.
A Japanese study released just days ago found that active workplace AI users reported efficiency gains nearly three times higher than non-users 58.7% versus 20.1%. These aren’t marginal improvements. These are people completing their work in a fraction of the time it used to take.
But here’s the kicker: Among those same active AI users, 52.9% expressed concern that AI could deteriorate human thinking capacity, and 36.5% worried it would make it harder to demonstrate their unique value. The tools making them more productive are simultaneously making them question their own worth.
Token Anxiety: The New Workplace Affliction
There’s a term that’s emerged in Silicon Valley for this phenomenon: token anxiety.
Nikunj Kothari, a venture capital investor in San Francisco, coined it to describe the obsessive behavior he’s seeing everywhere. People checking on their AI agents during parties. Laptops glowing softly in dark corners of bars. Engineers running automation scripts while supposedly “touching grass” in Dolores Park.
“Everybody has this feeling of like, ‘Hey, time is the only thing that matters. And in that given unit of time, which we don’t get back, how can I have AI do a lot more for me than the next person?’” Kothari explained.
It’s not about working smarter anymore. It’s about working non-stop, 24/7, using AI agents that never sleep. The new flex in tech isn’t “how big is your team,” it’s “how big is your agent swarm.”
Some companies are now tracking the number of times engineers interact with AI coding tools daily. The higher the number, the assumption goes, the more productive the team. Weekly reports analyze patterns where developers got stuck in ineffective loops with AI and provide suggestions for improvement.
One VP of Product admitted developing something close to addiction: “I feel like I have to complete several more interactions every day, and I’m still thinking about how to do a few more before I go to bed.”
This isn’t healthy. And everyone knows it. But the competitive pressure is overwhelming.
The Perception Gap That’s Driving Everyone Crazy
Here’s where things get really interesting and frustrating.
There’s a massive gap between how executives perceive AI’s benefits and how frontline employees experience them. Executives tend to see AI as a clear time-saver. Employees? Not so much.
In fact, some research shows that workers report AI is actually making them less productive, not more. Time spent on certain job responsibilities has increased by up to 346% as people learn to work alongside these tools. Email time has doubled. Focused work sessions have fallen by 9%.
Why the disconnect?
Partly, it’s because employees bear the transition costs. They’re expected to learn new tools, experiment with AI workflows, and figure out optimal prompting strategies all while their daily work expectations remain unchanged. Nobody’s giving them less work to make room for the learning curve.
But there’s something deeper happening too. When a marketer produces first drafts 40% faster using AI, they rarely get to enjoy that extra time. Instead, output expectations rise to fill the gap. Produce five blog posts instead of three. Create ten variations of that campaign instead of five. The efficiency gains get absorbed into higher workload demands rather than showing up as actual time saved.
Economists call this the productivity paradox. It’s the gap between how transformative a technology feels in use and how long it takes to appear in aggregate business results.
We saw the exact same pattern with personal computers in the 1980s and 90s. Companies bought PCs, saw no productivity improvement for nearly a decade, and then the gains suddenly materialized as workflows and organizational structures actually restructured around the technology.
We might be in that waiting period right now with AI.
The Fear Nobody Wants to Say Out Loud
Let’s address the elephant in the room: job displacement.
A survey of U.S. workers found that one in five expressed direct concern about displacement, noting that their job or at least aspects of it is being taken over by automation. Workers in jobs identified as most exposed to AI voiced worry three times as often as those in less at-risk positions.
And they have reason to worry. Companies that discussed AI in the context of their workforce reduced their job openings by 12% over the past year, compared to an 8% reduction across all companies. About 55,000 layoffs were attributed to AI in 2025, according to research firm Challenger, Gray & Christmas just 4.5% of all job losses, but a number that’s expected to increase by as much as 9x in 2026.
Goldman Sachs forecasts that 6% to 7% of workers roughly 11 million jobs will eventually be displaced by AI automation over the long term.
But here’s what makes this particularly cruel: the people most likely to be displaced are often the same people seeing the biggest immediate productivity gains.
Entry-level developers learning to code faster with AI assistance. Junior analysts processing data more efficiently. Customer service representatives resolving tickets quicker with AI-suggested responses.
These productivity improvements are real. But they’re also making companies ask: If this junior person can do the work of someone more senior using AI, do we still need to hire as many people at each level? And if we can train AI on what our best performers do, couldn’t we potentially replace more expensive labor with cheaper AI-assisted labor?
The Tasks That Are Disappearing
What’s getting automated isn’t always obvious. It’s not just repetitive grunt work (though that’s definitely going).
According to workplace researchers, the most exposed tasks include documentation, basic coding, routine analysis, and structured support work. These tasks often sit at the base of the experience ladder the very work that traditionally gave entry-level workers a way in.
As one analyst put it: “What you begin to lose is not the job. It is the path into the job.”
Think about it. If junior developers don’t spend time writing boilerplate code because AI handles it, they miss out on pattern recognition that builds intuition for architecture. If junior analysts don’t manually clean messy datasets, they don’t develop the critical thinking that leads to asking better questions.
Companies may not realize the delayed impact until years later when they discover they don’t have enough mid-level experts because they didn’t bring enough people in at lower levels.
It’s a systemic problem that nobody’s really solving yet.
The Rewards Problem
Here’s another tension point that’s brewing: compensation.
Workers are using AI to dramatically boost their productivity. They’re delivering more value in less time. But only 17% of workers report receiving tangible rewards raises, bonuses, promotions tied to AI-enhanced performance.
From an employee’s perspective, this creates a nasty dynamic. You learn to use AI tools (often on your own time, at your own expense 66% of workers are personally funding AI tools for work). You figure out how to integrate them into your workflow. You start delivering significantly more output.
And then… nothing changes. Your salary stays the same. Your title stays the same. The company captures all the value from your increased productivity while you just end up with more work.
One survey found that 45% of employees have used AI at work without informing their manager. Why the secrecy? Partly because they’re not sure if it’s allowed. But also because they’re not sure what happens if their manager realizes how much AI is doing.
Will they get credit for being innovative and efficient? Or will they be seen as redundant?
The Companies That Are Actually Figuring It Out
Not every organization is stuck in this anxious holding pattern. Some are starting to crack the code on making AI work for people, not against them.
The pattern among successful adopters is consistent: they’re not just buying software licenses and telling employees to “use AI more.” They’re implementing specific AI workflows for specific roles, with clear ownership and training.
One example: a legal team trained specifically to use AI for contract review, with defined protocols for when AI suggestions should be accepted, modified, or rejected. The team gets better, faster results. Individual lawyers don’t feel threatened because they understand exactly how the AI fits into their expertise rather than replaces it.
Another example from a benefits analyst who was drowning in repetitive questions. Working with an AI specialist, they built custom prompts aligned with tone requirements and legal considerations. Crucially, ownership of the solution stayed with the analyst. She became one of the company’s strongest AI advocates not because it threatened her job, but because it freed her to do the parts of her job that actually required human judgment.
Companies reporting the best results share another characteristic: psychological safety around AI discussions. They don’t avoid the hard questions about job security. They address concerns directly and involve employees in shaping how AI gets deployed.
As one HR leader put it: “The fear of being obsolete is real. Avoiding the question of whether AI might take someone’s job erodes trust.”
Why This Matters Beyond Individual Anxiety
This isn’t just about individual workers feeling stressed (though that matters too). This productivity paradox has broader economic implications.
If productivity genuinely surges but companies don’t hire more people or raise wages, where do those gains go? Historically, productivity improvements have been shared between workers (through higher wages) and companies (through higher margins). But if companies capture all the gains this time, we could see a fundamental shift in how income gets distributed.
Some economists worry that widespread AI displacement without adequate redistribution could create a demand shock. If lots of people are unemployed or underemployed, who’s buying all the products and services that these newly efficient companies are producing?
Others are more optimistic, pointing to historical precedent. When power tools were introduced in construction, they didn’t eliminate construction workers. They made each worker more productive, which lowered costs, which increased demand for construction, which ultimately employed more workers.
The question is whether AI follows the power tool pattern or breaks it.
The Skills That Still Matter
So if you’re reading this and feeling the anxiety yourself, what should you actually do?
First, recognize that the workers reporting the most positive relationships with AI aren’t those who use it to do their existing jobs faster. They’re the ones using it to expand into capabilities previously outside their competence.
In other words, if AI helps you do routine data analysis 50% faster, that feels like replacement. But if AI helps you build financial models you couldn’t have built before, or create data visualizations that were beyond your skill level, or conduct competitor research at a scale that wasn’t previously feasible that feels like expansion.
Tech leaders who want to reduce AI anxiety should design deployments around capability extensions, not speed improvements on existing tasks.
Second, focus on skills that AI currently struggles with:
- Complex judgment in ambiguous situations. AI is great when problems are well-defined. It’s still pretty bad when the question itself is unclear or when success depends on reading subtle social dynamics.
- Genuine creativity and taste. AI can generate variations on existing patterns. It’s much weaker at truly novel approaches or developing distinctive aesthetic sense. You know how you can usually tell when a logo was AI-generated because it has that slightly generic quality? That’s the gap human designers can still fill.
- Deep contextual understanding. AI tools lack institutional memory and cultural nuance. If your value comes from understanding how your specific organization works, knowing the unwritten rules, or reading between the lines of what stakeholders really want, that’s still largely protected.
- Synthesis across domains. AI is trained on siloed data. Humans who can connect insights from completely different fields applying biological principles to organizational design, or using game theory to solve marketing problems still have a major edge.
Third, get really good at working with AI rather than seeing it as competition. The future probably isn’t “humans vs. AI.” It’s “humans good at working with AI vs. humans not good at working with AI.”
The Shadow Economy of Stealth AI Use
Here’s something most companies don’t realize: there’s a massive shadow economy of AI use happening right under their noses.
Four in five workers now use AI at work. More than a third consider it essential to their job. But 45% have used AI without telling their managers.
People are quietly revolutionizing how work gets done, and management often has no idea it’s happening.
This creates some weird dynamics. Companies are benefiting from productivity gains they don’t even know they’re getting. Workers are taking initiative to improve their workflows but hiding it because they’re unsure if it’s allowed or worried about consequences.
There’s also a related problem: resume inflation. One in four workers has either exaggerated their AI capabilities in job applications or intentionally overstated their skills during hiring. They’re betting they can learn fast enough to justify their claims.
The market pressure to demonstrate AI competency is so intense that authenticity is taking a backseat to survival.
What Actually Needs to Change
If we’re going to move past this anxiety-productivity paradox, several things need to happen:
Companies need clear AI policies. The ambiguity is killing people. Workers need to know: Is AI use encouraged? Required? Restricted? What tools are approved? What tasks should use AI and which shouldn’t? Clarity reduces anxiety even when the answers aren’t what people might hope for.
Measurement needs to evolve. Tracking “number of AI interactions” or “time saved” captures the wrong things. Better metrics might include: quality of output, capability expansion, learning velocity, customer satisfaction when AI is involved, or employee well-being alongside productivity.
Training needs to be real. Sending everyone a link to ChatGPT and expecting magic to happen isn’t a strategy. Effective AI adoption requires role-specific training, supported experimentation time, and ongoing coaching.
The rewards need to be shared. If employees are driving genuine productivity gains through AI adoption, that should show up in compensation, career advancement, or at minimum, workload relief. The current pattern where companies capture all the value isn’t sustainable.
Career paths need reimagining. If entry-level tasks are being automated away, how do people develop expertise? Some companies are experimenting with “apprenticeship 2.0” models where junior people work alongside AI on increasingly complex problems rather than grinding through years of routine work.
The Uncomfortable Truth
Here’s what I think is actually happening, and it’s more nuanced than either the optimists or pessimists want to admit.
AI is genuinely making many workers significantly more productive at specific tasks. Those gains are real and measurable when you measure at the individual task level.
But those individual gains aren’t automatically translating to organizational-level productivity improvements for several reasons:
- Learning overhead. People are spending huge amounts of time learning to use AI effectively, time that doesn’t show up in productivity calculations but absolutely affects output.
- Task expansion. Freed-up time is getting filled with more tasks rather than showing up as efficiency gains in business metrics.
- Quality variance. AI output requires human review and refinement. Sometimes that’s quick; sometimes it’s surprisingly time-consuming.
- Integration challenges. Most organizations haven’t restructured workflows to actually take advantage of AI capabilities, so gains happen in isolated pockets rather than systematically.
- Misalignment of incentives. Workers are hesitant to fully leverage AI if they think it might eliminate their jobs, leading to suboptimal adoption patterns.
The people seeing the biggest gains and experiencing the most anxiety are the early adopters who’ve pushed past these barriers. They’ve figured out effective AI workflows. They’re seeing what’s possible.
And that’s exactly why they’re worried. They can see the trajectory. They know that what takes them 30% less time today might take 60% less time next year. They’re wondering when the efficiency improvements cross over into redundancy.
So What’s the Actual Takeaway?
If you’re someone benefiting from AI productivity gains right now, your anxiety isn’t irrational. You’re seeing something real. The technology is advancing fast, and nobody knows exactly how this plays out.
But anxiety without action is just suffering. Here’s what you can actually control:
- Become genuinely expert at AI-assisted work in your domain. Don’t just dabble. Get really, really good at it. The people who thrive will be those who develop sophisticated judgment about when and how to use AI, not those who avoid it or use it superficially.
- Document and communicate your expanded capabilities. If AI helps you tackle problems you couldn’t before, make sure that’s visible. Frame yourself as “person who can do X, Y, and Z with AI assistance” rather than “person who does traditional tasks faster.”
- Build relationships and context that AI can’t replicate. Invest in institutional knowledge, cross-functional relationships, and deep understanding of your organization’s specific challenges.
- Stay flexible. This is moving fast. The skills in demand next year might be different from today. Maintain adaptability rather than doubling down on a single expertise area.
- Take care of your mental health. Seriously. The “token anxiety” pattern of obsessive productivity monitoring isn’t sustainable. If you find yourself checking AI agents at midnight or feeling genuine addiction to productivity tools, that’s a warning sign.
The productivity gains are real. The anxiety is real too. Both can be true at the same time.
The winners in this transition won’t necessarily be the people who squeeze the most output from AI in the short term. They’ll be the people who figure out how to use these tools in ways that make them more valuable, more irreplaceable, and more human not less.
That’s a harder path than just running your agent swarm 24/7. But it’s probably the one that leads somewhere you’d actually want to be.
Leave a Reply