By Garbo Tian

The AI layoff wave (often manifested as the "630" downsizing) is an industry-wide efficiency restructuring where technology giants utilize generative AI to automate R&D, merge development functions, and aggressively eliminate high-earning talent. By analyzing first-hand accounts from tech veterans, this deep dive explores why senior programmers, top performers, and middle managers have become the primary casualties of the AI transition.

Interviews: Cairu Ren, Jie Lan, Qianwen Peng

Editor: Cairu Ren, Qian Qiao, Xuan Yang

Source: 36Kr Future Consumption

The "630" Downsizing: Is AI the Culprit or the Scapegoat?

"The company has a layoff list, and you are on it." Called into a conference room by his team leader on a mid-May day, Lin Yue was met with blunt reality.

Lin’s initial reaction was calm; he had seen it coming. Rumors of internal cuts at several tech firms had been circulating since early spring. Since the beginning of the year, major Chinese internet companies have aggressively pursued AI-driven efficiency—launching token-consumption races, mandatory training sessions, and stealth performance reviews. When an entire industry is swept up in an "all-in on AI" campaign, layoffs become an unspoken certainty.

Yet, standing outside the HR office, his composure cracked. His hands shook. He hesitated, rehearsing his opening line and trying to manage his facial expressions. "I never want to go through this again."

Earning a monthly salary of 25,000 RMB, Lin joined Trip.com as a backend engineer straight out of college a year ago—a seemingly immense stroke of luck. As the tech hiring boom cooled, he was one of fewer than 500 applicants hired from thousands of resumes, landing in the company's highly profitable hotel division to write code for commercial products.

In hindsight, however, a junior programmer with a high salary and barely a year of experience is the perfect target for elimination. First, severance costs are minimal. Second, compared to seasoned veterans familiar with overarching business logic, junior staff leverage AI less efficiently. "With a solid foundation in business operations, older employees know exactly what they want AI to do and how it impacts the bottom line," Lin explained.

A Stanford University paper, Canaries in the Coal Mine?, likens young entrants in the workforce to the proverbial canaries. The research indicates that since the widespread adoption of ChatGPT in 2022, employment among the youngest demographic has dropped sharply. Projections suggest that by September 2025, employment for software developers aged 22–25 will plunge nearly 20% from its late-2022 peak.

Over the past year, AI has intensified the grind. Trip.com was once colloquially known as a "retirement home" within the internet sector: programmers clocked in at 10:30 AM, took two-hour lunches, and left precisely at 7:00 PM, with the main app undergoing bi-weekly iterations. Shortly after Lin onboarded, however, the explosion of AI coding capabilities accelerated the pace to weekly iterations. "We work until 10:30 PM every day."

This accelerated tempo is not driven by explosive business growth. "If we don't find things to do, we become a marginal department, and marginal departments get cut," Lin said. Ultimately, he could not escape that fate.

The cuts, however, can be entirely indiscriminate.

Cang Shu never expected to be among the first on the chopping block. On a Friday morning in May, half an hour before his shift, "the department suddenly pulled everyone into an all-hands meeting, and HR simply announced the outcome."

Before joining Meituan, Cang was a Super Special Offer (SSP) campus recruit at ByteDance, securing a premium starting salary that left him the highest earner among his peers. After jumping to Meituan, he handled almost all core projects within his group. This year was supposed to mark his promotion.

In this wave of layoffs, the protective shields of "top performer" and "senior rank" collapsed. In the adjacent team, two dismissed employees had received "exceeds expectations" performance ratings the previous year. Eventually, Cang’s entire unit was effectively eradicated. "The team exists in name only; practically no one is left."

When Lin learned of his dismissal, he realized the profile pictures of two frontend engineers he frequently collaborated with "had already turned gray." A Meituan user-growth chat group, once boasting hundreds of members, now retains only half. Alibaba’s Amap and Fliggy divisions are experiencing similarly violent tremors.

"630" (June 30) has become a trending buzzword on social media, marking the end of the first quarter where AI truly infiltrated the Chinese internet workplace at scale. Late June to mid-July is a traditional window for corporate turnover, but this time, it serves as the universal "last day" across the current layoff cycle.

Silicon Valley, the industry bellwether, preempted this trend with massive, batch-oriented layoffs. In May, Meta cut 8,000 jobs while transferring 7,000 employees to its AI division, rendering it the most volatile tech firm in the Valley. Executives conceded that "company morale is at a 20-year low." Earlier, Amazon eliminated 16,000 white-collar roles to redirect capital toward AI.

Prior to the previous layoff wave in 2021, major Chinese tech companies maniacally expanded their boundaries, launching new ventures at high density. Armies of personnel were rapidly recruited and just as rapidly purged.

This year’s narrative is far more complex. AI-driven efficiency mandates, stagnant legacy operations bogged down in competitive quagmires, and the cash drain of funding new AI ventures are inextricably intertwined. Many of those shown the door struggle to weigh which factor dealt the final blow.

As the author of Demis Hassabis: The Brain Behind Google's AI observed, much like J. Robert Oppenheimer, who birthed the atomic bomb but could not control its deployment, scientists pursuing truth are also "destroyers of worlds." Our jobs, cognitive frameworks, and very survival face potential disruption. A decade ago in Seoul, AlphaGo delivered the first shock to human Go player Lee Sedol. Ten years later, from Silicon Valley to Beijing, that disruption is metastasizing.

For large corporations, AI is the boarding pass to the future, funneling resources toward large language models and native applications. Yet, no one can predict if or when these nascent ventures will succeed. Confronted with flatlining legacy businesses, tech giants are forced to decisively drive efficiency—and consequently, execute layoffs—across both certain and uncertain trajectories.

Confiding in a friend, Lin was consoled: "It's okay. We will all face this day; yours just arrived earlier." But more critical than self-consolation is the question of what comes next. Once replaced by AI and expelled by Big Tech, how should people adapt and move forward?

Anxious Executives, Escalating Middle Management, and Frenzied Grassroots

"Product demos that took two months to build at ByteDance can now be spun up in two weeks," a former ByteDance product manager and current AI startup executive explained. Armed with tools like Claude Code and Codex, his team now drafts demos in three hours and validates concepts within a week.

"A product manager now operates like a CEO," he added. Organizational structures shrink accordingly, cutting the frictional costs of communication common in large corporations—achieving perfect "entropy reduction."

While startups pivot rapidly via AI, do internet behemoths look inward and see sluggish leviathans? Statements from the highest echelons of Big Tech often serve as the definitive signal.

During an executive briefing this March, Meituan CEO Wang Xing shared his perspective: "AI Agents impact me more profoundly than ChatGPT. AI will inevitably generate immense productivity and force massive structural shifts in organizations and work models."

Shortly after, Meituan convened a company-wide virtual town hall to mandate the installation and utilization of "Lobster" (an internal AI tool), urging every employee to integrate it and codify their daily routines into reusable "Skills."

Following the meeting, Chen Yujia, a Meituan merchant operator in the core local commerce division, was instructed to add a dedicated section to her weekly report detailing her AI-driven efficiency gains and proposing "Skills" for team and departmental adoption. "You could feel everyone desperately trying to shoehorn AI into their workflows."

In April, an Alibaba algorithm engineer abruptly received the department's token consumption leaderboard for the previous month. He was publicly commended for topping the chart with 17 billion tokens. Department heads indicated that future annual KPIs and promotion cycles would factor in this ranking. A month later, however, the updated leaderboard never materialized—management likely realized the metric was flawed.

New directives followed swiftly. Department leaders demanded hourly "time reports" between 11:00 AM and 6:00 PM, utilizing Agent plugins to auto-log code and dialogue histories to generate work summaries—meaning employees could not alter their own logs. By the very next day, HR violently opposed and dismantled the absurd policy.

Such incidents are no longer shocking. Executive AI anxiety cascades downward. Middle managers escalate the pressure, implicitly signaling to subordinates that they are trapped in a shadow race of reporting, arming, and surviving.

Though authoring "Skills" wasn't explicitly mandatory, Chen's manager closely monitored token usage, frequently probing for details. "He doesn't fully understand what AI can do either, but he stated point-blank that no one on his team is allowed to fall behind in this AI wave." During private post-work dinners, the unspoken threat lingered: "You must utilize AI; otherwise, I won't be able to save you when the time comes."

An engineer working on an Alibaba AI coding product revealed that business unit heads requested data tracking features "to gain clear visibility into the daily AI usage trajectories of their team members."

AI-driven efficiency has mutated into a checkbox for every business unit and function. Yet, a massive chasm separates the grassroots from management regarding AI's actual utility and implementation. Executives project infinite optimism; rank-and-file employees exhaust themselves trying to materialize those visions, ultimately settling for exhausting "performative" productivity.

Jiang Ling handles customer operations at Alibaba's Taotian Group, matching consumer demand with merchant supply. In her view, management consistently "assumes AI is exceptionally intelligent and simple to execute."

Take "order explosions"—a common e-commerce anomaly. Executives expect comprehensive AI patrols to preemptively flag all potential viral products. However, the platform hosts tens of millions of items daily, drastically exceeding current human and token bandwidth. Consequently, testing is restricted to small-scale samples of a few hundred thousand items, yielding abysmal hit rates.

"As an employee, you simply cannot argue against the boss's expectations, you know?" Jiang expressed, caught between outrage and resignation. Often, she feels like a donkey chasing a carrot, spurred by a whip. "Exhaustion isn't the terrifying part; operating without direction or positive feedback is. You keep grinding the mill without knowing the destination."

"You cannot treat AI like a wishing well," summarized the CTO of an AI startup. Efficiency gains require prerequisites, primarily clean data—a foundational element many companies lack. Furthermore, operational bottlenecks are fundamentally "human" problems that AI alone cannot solve.

Every Generation Faces Its Own Sunset Industry

While product managers and operators grapple with uncertain anxiety, programmers are already absorbing the impact of a definitive sentence.

Li Chuan, a Baidu frontend engineer, was first stunned by AI earlier this year when he utilized Claude Code. "Complex requirements that demanded five or six prompt iterations with domestic models were resolved flawlessly by Claude in two or three."

His second shock came in April when Zhipu AI released the GLM-5.1 model. "It was highly cost-effective and fully capable of serving as a Claude Code substitute." Li immediately realized his job was in jeopardy. By May, his name unsurprisingly materialized on the "list."

Two sides of a coin: on one side, by May 2026, Claude Code’s parent company, Anthropic, hit an Annual Recurring Revenue (ARR) of approximately $47 billion, expanding four to five times over a six-month period; Zhipu also recently surged to a trillion-dollar valuation. On the flip side, the rapid maturation of AI coding capabilities has transformed programmers into the primary casualties of this layoff cycle.

"Almost every company takes an axe to its R&D teams first, particularly frontend and test development roles, which executives increasingly view as low-value," an internet company HR representative noted.

In 2025, Li entered Baidu as a campus recruit. During his interviews just a year prior, AI functioned merely as a search engine and basic coding assistant. His interviewers never once broached the topic.

Frontend engineering was Li’s dream job because of its WYSIWYG (what-you-see-is-what-you-get) nature; code quality translates directly to UI details. Every Lunar New Year, telling his family, "Open the Baidu app, I built that element right there," gave him a profound sense of achievement and purpose.

For years, big tech programmers were strictly siloed into algorithm, frontend, backend, and testing functions. Frontend required soft skills like aesthetics and interaction design, while backend demanded rigorous technical logic. Salary bands and the internal "chain of disdain" were directly tied to this perceived technical depth—frontend engineers out-earned testers but lagged behind algorithm and backend engineers.

In just one year, Li’s familiar world inverted. Code generation and debugging have been heavily automated, blurring the rigid boundaries between programming functions. Even product managers can now dip their toes into coding.

An Alibaba development unit was notified in May to suspend all non-urgent requirements. Each team was tasked with building an Agent; moving forward, product managers would interface directly with the Agent for all business requirements. Programmers were relegated strictly to Agent maintenance, barred from writing direct code. Management implied that by October, high-performing teams would absorb the Agent maintenance duties of underperforming ones.

Tencent’s CSIG tech team developed a bug-fixing pipeline for the company app—AI resolves the bugs, and programmers merely review the fix. Upon clicking "confirm," the code merges. Its repair accuracy currently sits at 50%.

In May, Alibaba internally launched several full-stack teams, converting frontend, backend, and testing engineers into "full-stack engineers" or "super individuals." Starting in June, Meituan broadly pushed to merge frontend and backend development.

Transitioning to "full-stack" works in theory, but in practice, it is an excruciating process. Abruptly reclassified as a full-stack engineer, Han Zhi had zero time to upskill before tackling her first full-stack project, shouldering frontend, backend, and testing solo. "All my requirements are now reverse-scheduled with hard launch deadlines." Pushed to her absolute physical limits, she frequently finds herself working past 9:00 PM. "I am just too exhausted."

But the macro trend is irreversible. From late last year to early this year, top Chinese tech firms indiscriminately burned cash to force token consumption and phase out "manual coding."

At its peak, Tencent CSIG team members were allocated a monthly token quota of $2,000. Assuming reasonable requests and valid code output, they could request limit doubles upon depletion. Token consumption was hardwired into performance metrics. "If your usage drops, your leader demands an explanation." Consequently, some engineers loaned their surplus quotas to peers.

For years, big tech programmers enjoyed elite salaries and prestige. They were the bedrock of internet companies. The ethos of "programmer spirit" meant open-source sharing, elegant code, pure meritocracy devoid of office politics, and the raw thrill of watching characters execute on a screen.

Times have changed. Nearly every programmer interviewed echoed the same sentiment: "Working without AI is impossible. If the AI crashes, I'd rather spend hours searching for a new coding solution than manually write the code myself." Discussing the traditional "programmer spirit" now feels obsolete.

A top-tier programmer's hallmark used to be continuous learning, Li noted. Programming languages evolved constantly; stagnation meant obsolescence. Weekend coffee shop sessions with peers to master new stacks were routine. "The community itself is inherently competitive." Yet, AI's terrifying iteration speed has rendered them speechless.

"If AI coding had just plateaued at its 2025 levels, that would be ideal. It levels the playing field between someone with two years of experience and someone with eight, without fully replacing humans, leaving plenty of work outside the 'chat window'," Lin lamented. But technology bows to no one. He has zero doubts that the extinction of the programmer is currently underway—"just like textile workers after the invention of the spinning jenny."

Old Growth is Dead; the New Horse Race Begins

When technology injects an exponential efficiency lever into a corporation, only two outcomes emerge: the same headcount executes vastly more work, or the company no longer needs that headcount.

"We don't lay off," the CEO of a software firm claimed. Having finally trained programmers with deep industry context and development methodologies, he views each as a corporate asset. When AI coding boosts efficiency fivefold, his strategy isn't to fire 80% of the staff, but to scale the business fivefold.

The sentiment is idealistic, but the critical flaw remains: does the market actually possess that much untapped growth?

Prior to his termination, Lin briefly tasted the "liberation" of AI code generation. Quickly, however, his workload multiplied. Previously, business units tolerated slow scheduling for minor app iterations. Now, demands pile up rapidly. Regardless of feasibility or strategic importance, R&D is told to "just build it and test it."

To Lin, these tasks felt marginal—tweaking copy on minor banner ads, or switching a pop-up prompt from "free cancellation" to "points deduction." "Product managers tinker endlessly, running A/B tests, but the post-modification metrics rarely improve significantly."

"The departments with the least growth lean hardest into AI. They need a new narrative to pitch," Cang noted. Having worked in both food delivery and drone units, he observed firsthand that the former pushes AI far more aggressively than the latter.

An infrastructure engineer who just survived Meta’s mass layoffs shared that after learning to squeeze productivity out of AI, colleagues wanted to experiment with projects they previously lacked time for. But as headcount drops, the remaining staff invariably chop those low-priority tasks back off the ledger.

The stark reality facing the industry is that the star products of the mobile internet era can no longer generate material growth simply by "doing more." Many are flatlining, bleeding capital amid vicious external competition.

The 2025 food delivery wars burned 200 billion RMB across several players, dragging Meituan's margins and cash flow into the mud. Consequently, Meituan—historically plagued by low per-capita profit contributions—entered the layoff cycle first. Conversely, Meituan is highly dependent on offline fulfillment, meaning its AI efficiency ceiling is lower than fully digital peers. "If even Meituan can shrink headcount via AI, other firms will absolutely follow suit. It’s a bellwether," a Meituan employee remarked.

Baidu, wrestling with shrinking legacy ad revenues, and Alibaba's perpetually marginal Fliggy and Amap units, face identical dynamics.

If cuts in legacy divisions are inevitable, do internal mobility opportunities still exist?

Discussing the cuts, some managers advise employees to "look for AI-related projects within the company." Meituan's core local commerce unit recently spun up an AI Transformation department focused on streamlining internal workflows. Furthermore, several senior executives are directly spearheading AI initiatives.

Wang Yue, a ByteDance product manager, is operating an internal startup targeting B2B AI efficiency products. "The company actively encourages this exact kind of exploration." At inception, the team preemptively eliminated "design" and "testing" roles, pitching the review board solely on future labor cost reductions. Another colleague developing an AI customer service Agent set a 2026 OKR to "help the company lay off XX% of customer service reps."

Today, dozens of these strike teams operate within every major tech firm. "Multiple teams often chase the identical vertical; whoever gains traction gets the consolidated corporate resources." A new horse race has commenced.

Beyond shifting business priorities, organizational topologies are flattening, aggressively erasing middle management. Starting this year, Tencent pivoted to a project-based structure, diluting managerial hierarchies and reinstating technical tracks for leads. During mid-year reviews, Meituan purged several L9 (Business Unit Director level) executives and completely abolished the X1 node (the lowest managerial tier) to compress reporting lines.

Bidding Farewell to the Past

Few have experienced an "epiphany" regarding where this AI tsunami will ultimately deposit human capital.

Approaching the mid-June end of his transition period, Lin was aggressively interviewing at Taobao, Kuaishou, and ByteDance. Prolonging his Big Tech career remains his optimal mental blueprint. But the offers haven't materialized. "It’s too difficult," he admitted.

"Finding a job is easy. But once you downgrade from a Tier 1 giant to a mid-size or small firm, you can never climb back." To Lin, surrendering the Big Tech halo represents a permanent demotion he refuses to accept.

Others have shed their Big Tech obsession. Three days after exiting Baidu, Li Chuan seamlessly onboarded at a startup, naturally transitioning from frontend to full-stack. Building office productivity AI Agents, the startup even offered him a pay raise.

Despite the consensus that the era has shifted and programming skills are no longer reliable, Li retains technical aspirations. He wants to code a product users genuinely love—a goal not exclusive to massive corporations.

Post-Alibaba, Jiang Ling joined a legacy automaker. Her new role requires zero forced AI integration. She no longer sweats over impossible management AI KPIs or engages in performative exhaustion. Her current project launches on September 30. "These tasks sit comfortably within my strike zone, and the timeline is generous. The mental and physical relief is immense." Recently, whenever her department posts an opening, "floods of Alibaba alumni show up, desperately rushing into manufacturing."

Perhaps 10% of programmers will survive the purge, but Cang refuses to job-hunt in Big Tech just to "grind for a spot in that desperate 10%." Fired from Meituan in May, he decisively pivoted to entrepreneurship. Even before the AI wave, he hustled side projects. Building communities and monetizing skills previously netted him 100,000 RMB a month.

By March and April, members of his community were already surfing the AI wave to launch startups. "They founded companies and hired aggressively, while I was still grinding away at my corporate job. Did that make sense?" he asked himself.

Today, Cang’s startup targets overseas markets, developing systems and standalone products for users with rare diseases. He documents his journey on social media under the handle "Cang Shu (Salary-Quitting Edition)." He runs multiple side products simultaneously to stay sharp. "A small tool takes three to four days max; a complex system might take half a month"—lightyears ahead of Big Tech's standard scheduling.

AI is arguably the most potent cognitive lever in human history. It amplifies individual capability exponentially, accelerates startup execution, and ensures brilliant ideas are rapidly visualized and monetized.

Born in 2000, Cang insists he was destined for entrepreneurship, but without the layoff, he might not have pulled the trigger right now. "The company made the decision for me."

"Hold no attachments to the past; push relentlessly forward." This is the closing line in Meituan's automated farewell text to departing employees—a phrase currently echoing among the newly unemployed. Amid this complex AI-driven paradigm shift, whether exiting or remaining in Big Tech, legacy playbooks are obsolete.

A brief moment of shattering does not equate to defeat. Whether pivoting industries or launching startups, those who internalize the shift first will likely map the new terrain before anyone else.

(Xinyu Zhou contributed to this article. At the request of the interviewees, Lin Yue, Jiang Ling, Li Chuan, and Wang Yue are pseudonyms.)