The more powerful AI becomes, the more tired people become, and "anxiety" becomes the norm for both companies and employees.

Author: Xu Chao , Wall Street Insights

AI programming tools promised to liberate engineers, but in reality, they have given rise to a new wave of efficiency anxiety.

As the capabilities of AI programming agents such as Anthropic's Claude Code and OpenAI's Codex continue to leap forward, tech companies are caught in a top-down "productivity obsession." Executives are personally writing code, employees are being asked to increase the frequency of their interactions with AI, and overtime hours are increasing rather than decreasing. AI, which should be a labor-saving tool, has become a new source of stress in many workplaces.

Survey data reveals a significant cognitive gap: a survey by consulting firm Section shows that over 40% of C-level executives believe AI tools save them at least 8 hours per week, while 67% of non-management employees say AI saves them less than two hours, or even none at all. An ongoing study by UC Berkeley of a 200-person organization found that even with employees offloading a significant amount of work to AI, actual working hours are still increasing.

This spread of anxiety has structural causes. When CTOs are coding in front of AI at 5 a.m. and CEOs are measuring team effort by bill amounts, the entire industry's perception of "efficiency" has been redefined—and the cost of this redefinition is being borne by ordinary employees.

Executives are getting involved in coding, and efficiency anxiety is spreading from top to bottom.

The term "Vibe coding" initially carried a sense of languid anticipation. Former OpenAI researcher Andrej Karpathy introduced the concept to the public in February 2025, describing a new programming paradigm where engineers could complete development simply by chatting with AI—"completely immersed in the atmosphere."

However, a year later, the atmosphere had already changed.

Intuit CTO Alex Balazs described his recent routine: his wife came downstairs at 8 a.m. and found him already working for hours. "She asked me how long I'd been up, and I said I'd been writing code since 5 a.m." To be precise, he was guiding an AI agent to write code for him, which he said allowed him to reconnect with the underlying code he hadn't touched in years.

This kind of executive behavior is transmitting pressure downwards. Greg Brockman, president of OpenAI, recently posted on X, "Every moment your AI isn't running feels like a wasted opportunity." This statement precisely addresses the workaholic culture that is already prevalent in the tech industry.

Alex Salazar, co-founder and CEO of AI startup Arcade.dev, takes a more direct approach. He regularly reviews the company's Claude Code bills—the amount of which is directly linked to how often engineers use the tools—and singles out employees who "don't spend enough": "I'd say, 'You're not working hard enough.'" He says that after the first such "faith meeting," the company's AI programming tool bills skyrocketed tenfold, which he sees as a sign of progress.

As employees are subjected to quantitative management, "AI fatigue" is quietly spreading.

In this atmosphere, the way employees are evaluated is also quietly changing.

DocuSketch, a software company specializing in property repair, now tracks the number of times engineers interact with its AI programming tools daily, according to Andrew Wirick, VP of Product. The higher this number, the more productive the team is, by default. Claude Code also generates weekly reports for each engineer, listing all patterns where they've gotten stuck in ineffective loops with the AI ​​and providing improvement suggestions.

Wirick himself admitted that he has developed a kind of "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." He attributed this state to the "epiphany experience" he had when he tried Anthropic's latest model Opus 4.5 last November. At that time, he gave the model a functional prototype task that would normally be handed over to an engineer. After 20 minutes, he saw the model autonomously disassemble and implement the task, "It felt like my brain was rebooted".

This mindset of everyone speeding up is eroding the boundaries between work and life. Research from Berkeley found that even though AI has taken over a large number of tasks, people's working hours have not decreased. Some engineers have also begun to openly admit that they are experiencing "AI fatigue"—a constant worry about missing the next breakthrough, which always seems to be just one cue word away.

The cognitive gap between executives and employees is widening.

The executives' enthusiasm largely stems from the novelty of creating things themselves. Salazar admits that building prototypes with AI himself is more "productive" than handling delegation and decision-making in his daily work. He even recently responded directly to a service request from a major financial client, building a demo application from scratch.

At Intuit, product managers and designers are now encouraged to build feature prototypes themselves in QuickBooks using "vibe coding." Balazs says, "At least now, product managers can take something concrete to the engineers and say, 'I want something like this.'"

However, data from a survey by Section Consulting shows that this cognitive gap is quite significant.

There is a significant gap between how executives perceive the benefits of AI and how frontline employees experience it. Salazar believes this is partly due to the higher transition costs employees bear when adapting to new tools: "They are implicitly required to find time to explore and experiment, but their daily work expectations are not adjusted accordingly to make room for this."

Concerns about job security are also a real concern. Salazar admitted that he had originally planned to switch to a third-party web service provider, but now that the marketing team can update the company website using AI tools, this outsourcing expense has been cut.

"Task Expansion" and False Prosperity: The Other Side of the Efficiency Myth

Berkeley researchers have termed this phenomenon "task expansion": when non-technical colleagues start using AI to generate code, engineers have to spend time cleaning up these unfinished products, which actually increases their workload. Intuit's Balazs admits that this is reshaping the originally clearly defined division of labor, making more and more roles "hybrid," and making existing collaborative relationships more complex.

The deeper question is: is this construction boom creating something of value, or is it just creating more things?

Analysts point out that if this AI-driven productivity obsession goes unchecked, it could lead to a surge of "busyware"—minor website tweaks that nobody cares about, custom dashboards for a single user, and prototype projects abandoned halfway through by marketing executives, all ultimately left to engineers to implement. Each of these may seem justified at the moment, but most will eventually end up in the dumping grounds of obsolete code.

According to Balazs of Intuit, the company's engineers have seen a roughly 30% increase in productivity, measured by code production and delivery speed. But in this future where code is becoming increasingly "disposable," the real efficiency gains may lie in the answer to another question: what things should never have been built in the first place.