The 35-Year-Old Programmer Crisis Just Got Rewritten by AI

After 35, the value of a programmer is being repriced

1. The old algorithm

For more than a decade, China’s internet industry ran on an age algorithm.

Once a programmer reached 35, if he had not moved into management, become an architect, or reached a high enough level at a big tech company, he was assumed to be losing value. The logic was not complicated: younger people have more energy, learn new technology faster, cost less, carry fewer family burdens, and their code does not look visibly worse than that of older employees.

In the years of rapid business expansion, most jobs did not require deep technical judgment. They required translating requirements into code quickly. A new graduate could be trained for half a year and then take over a large amount of routine development. A five-year veteran doing the same work would see his experience premium flattened fast.

The 35-year-old crisis did not arrive on one particular day. It accumulated: salary rising, energy falling, the tech stack overlapping more and more with that of newcomers, while the company’s ruler for measuring you was still “Can you write code like a 25-year-old?”

The pothos on the desk was replaced twice. The keyboard cover wore through three times. The number on the pay stub kept rising. The age range on the job posting kept tightening.

2. AI has made “hand speed” cheap

Over the past two years, the spread of vibe coding and AI programming tools changed something fundamental: translating requirements into code is no longer scarce.

With AI, a newcomer can build a working page, a simple backend API, or a data-cleaning script in a few hours. Work that once required reading docs, fixing environments, and debugging repeatedly can now produce a rough version from a single prompt. Coding speed has been flattened. Familiarity with a tech stack is no longer an absolute barrier.

The problem sits right there.

One frontline engineer’s observation is representative: new graduates who joined in the past two years are growing in a different direction. They turn earlier toward product thinking and architecture thinking. They are better at describing requirements in natural language and more dependent on AI for implementation. On the surface, they grow faster, skipping a lot of dirty work. But once they hit the hard 20%, which includes performance bottlenecks, concurrency conflicts, boundary conditions, system consistency, and online failures, they often lack enough technical detail to support judgment.

A pattern appears: newcomers and veterans have similar efficiency on 80% of problems. On the remaining 20%, the efficiency gap can reach dozens of times. Newcomers easily fall into loops with AI. AI gives a plan, it fails, AI gives another, it fails again. Veterans, in architecture design, solution discussions, and failure investigation, use memories of past pits to identify quickly that “this path probably will not work.”

The biggest change for veterans is that their ability to wrangle in natural language has shifted without friction into wrangling with AI. They know how to follow up. They know where AI tends to talk nonsense. They know which solutions look beautiful but break in production. AI covers their shortage of energy and their reluctance to write repetitive code. They cover AI’s lack of real-scenario judgment.

A judgment that sounds backwards is emerging: with AI, the advantage of seniority has grown, and the disadvantage of age has shrunk.

Add a qualifier: the beneficiaries are senior programmers with complex-system experience, not everyone who is older. Age is only a rough proxy for experience. What appreciates is the part of experience that is hard to standardize and can be turned into judgment and action.

If you have spent the past ten years mainly “translating requirements documents into code,” AI will replace you first. The essence of that work is standardized conversion. AI does it faster, cheaper, and without fatigue.

If you have spent the past ten years across different projects, failures, and business scenarios, building a feel for “why systems crash, why projects spin out of control, why solutions fail in practice,” AI becomes an advantage. You no longer need to write every line yourself, but you need to judge: what is worth doing, what is not, where problems will appear, which solution can survive longer.

3. The divide: three layers of ability

The abilities that avoid elimination can be split into three layers.

At the bottom is system-level judgment. AI can cover a small project, but once a project grows, the question becomes “Can this system survive?” How to split modules, where to set boundaries, how much extensibility to leave, how to achieve high availability, how to guarantee data consistency. AI is good at local completion, not at maintaining consistency over long periods under complex constraints. People who have never lived through a complex system losing control often cannot see when AI starts talking nonsense.

In the middle is the ability to “complete the last 1%.” AI can bring something to 80 points. The road from 80 to 95 requires deep understanding of the business scenario, intuition about risk boundaries, and trade-offs between cost and benefit. That 1% often decides whether a project can launch, or whether it can survive.

At the outer layer is breadth. People who understand code, understand business, can communicate, and are willing to face customers are scarcer in the AI era. AI can help a person go deep in one direction, but it cannot help a person judge “Is this worth doing at all? What does the customer actually want?” The higher the cross-domain degree, the less sufficient AI’s local completion becomes, and the more important human judgment is. Forward Deployed Engineers are in demand because the role crosses domains, and crossing domains is where AI helps least.

Conversely, the older programmers most easily marginalized are those who spent the past ten years mainly “translating requirements.” AI will not automatically turn you into a senior expert. It only amplifies your existing value, or accelerates the exposure of your replaceability.

4. The way out

For older programmers with average ability and average résumés, the way out is not to wait for AI to turn them into experts. Changing how value is measured is more realistic.

The first path is to move toward Forward Deployed Engineer. Demand for this role grew 42-fold from 2023 to 2025. ByteDance, Ant Group, Zhipu, and others are recruiting at salaries above comparable R&D roles. The core requirement is not technical depth, but understanding code, understanding business, being able to communicate, and being willing to go to customer sites. The job is translating a boss’s sentence “Can we use AI to improve efficiency?” into a deliverable plan. One practitioner described it this way: strong engineers often do not want long-term travel, while business-oriented people cannot handle model and system problems. This middle crossover zone happens to be the advantage zone for older programmers.

The second path is to go deep in a niche. Database optimization, performance tuning, security auditing, industrial software, embedded development. These fields have steep experience curves, and young people cannot catch up in a short time. Do not chase full stack. Pick a niche and go deep until “when people hit a problem, they think of you.” In reality, there is the case of a 42-year-old Linux kernel engineer fought over by three big tech companies with annual salaries of more than 2 million yuan.

The third path is to turn technical ability into service ability farther from code. Technical training, technical consulting, independent development, industry solution delivery. These directions do not demand high coding speed, but they demand high “knowing how to get things done.” A 55-year-old programmer, after being laid off, joined an AI company within a month, relying on a value position of “calculating costs clearly, being able to deliver, and being strong in both software and hardware.”

The fourth path is to accept an IT department in a traditional enterprise. Salaries often halve, but stability is relatively better. Technical roles in state-owned enterprises, public institutions, and traditional companies tolerate age more and demand less technical depth. The cost is a redefinition of income, pace, and identity.

5. After the layoff

Destinations are more concrete than paths out, and more uneven.

The more respectable destinations include: IT departments in traditional enterprises, technical roles in state-owned enterprises or public institutions, technical training and education, and remote or overseas outsourcing. A considerable number also turn to independent development or small technical consulting, using industry contacts built over the years to take on projects.

Less respectable cases exist and should not be avoided. One person, after being laid off, tried “selling sweet potatoes plus taking outsourcing gigs” and made 20,000 to 30,000 yuan in more than half a year. Another moved from a big-tech technical lead to system maintenance in a traditional enterprise, with monthly salary dropping from 62,000 yuan to 28,000 yuan. In the United States, there have also been cases of software engineers laid off and turning to welding or food delivery. Salaries dropped sharply, but the person said “a heavy stone had finally lifted from his chest.”

The common point in these cases: re-employment for older programmers often comes with a substantial drop in salary and a redefinition of identity. From architect to traditional-enterprise IT maintenance, from big-tech high-P to freelancer, income may be only one-third to one-half of the original. Those who transition smoothly are usually not the strongest technically, but those who first accept “defining their value in a different way.”

6. Big-tech credentials are not protection

Many people assume that becoming an architect, entering management, or getting big-tech credentials will almost certainly protect them from being eliminated by the industry. Even if they lose their jobs, that résumé will help them find work easily.

That judgment no longer holds in 2026.

Big-tech layoffs are moving toward middle managers. Tencent, JD.com, and other big tech companies are pushing “de-layering.” Among disclosed eliminated positions, more than 78% were concentrated in middle management at L5 to L7. The traditional protection of “high performer” and “high-P” no longer works. One employee said layoff intensity is directly tied to salary level: “The higher the salary, the more dangerous. Even ranking first in performance does not guarantee immunity.” The coordination and management function that middle managers once performed through information asymmetry is being weakened quickly by AI and flatter organizational structures.

The situation of full-time architects is also changing. Those who do not write code, do not carry business metrics, and only handle planning and design are hit first during industry contraction. A 39-year-old architect in Chengdu, after losing his job, sent out more than a hundred résumés and got zero interviews. His wife was unemployed, and he had two children. The pressure was pushed to the surface. A structural problem for architecture roles is that each project needs only one architect, while overall market demand for such roles is limited, unless a company happens to have an architect leave and open a slot.

Big-tech credentials carry less weight. When large numbers of laid-off big-tech employees enter the market at the same time, “ByteDance / Alibaba / Tencent background” turns from a screening advantage into a homogenized label. Hiring now rewards what kinds of problems you solved under what constraints, not which company you stayed at.

A person who made architecture trade-offs on a system with tens of millions of daily active users, and a person who maintained a small company’s business backend for five years: big-tech credentials will not make the former safer, but the latter’s depth of experience does form a defensible advantage.

7. Three variables

Breaking the age question into three independent variables is more useful than asking “Do older programmers have a way out?”

Variable one: Is your experience “standardizable” or “judgment-dependent”? Standardizable experience is being absorbed quickly by AI. Judgment-dependent experience appreciates in the AI era. Solutions highly dependent on application scenarios and impossible to package uniformly into skills belong to the latter.

Variable two: How much knowledge from different domains do you need to mobilize to solve a problem? The higher the cross-domain degree, the less sufficient AI’s local completion becomes, and the more important human judgment is. This is also why demand for Forward Deployed Engineers has surged. It is a role with a high cross-domain degree.

Variable three: Are you willing to accept a change in how value is defined? From “writing the best code” to “judging what is worth doing and what is not”; from “delivering at a desk” to “solving problems at a customer site”; from “big-tech high-P” to “a person who can calculate a deal clearly.” This shift is harder for many people than a technical transition, but it may be the most worthwhile investment after 35.

8. 65

The retirement age is 65. A 35-year-old programmer may be only halfway through a career.

Using a 35-year-old’s script to play out a life that runs to 65 guarantees anxiety. That script assumes you must complete every leap before 35, or you will be eliminated. AI is rewriting the script.

It makes coding speed cheap and judgment expensive. It compresses the growth path of newcomers and gives veterans’ memories of past pits a new use. It makes “senior” no longer just a synonym for high salary, but possibly an irreplaceable stabilizer in real scenarios.

AI has not made experience appreciate universally. It has accelerated the depreciation of “experience without judgment.” People who turned time into judgment are rebounding. Age itself is not.

For an individual, the realistic move is to redefine value early: turn experience into judgment, and judgment into deliverable value. Waiting for the industry to improve or trusting a résumé to keep you safe is not a plan. You can keep writing code, but do not only write code. You can keep doing technology, but do not only do technical implementation. You can stay at a big tech company, but do not depend only on a big tech company.

The age range on the job posting still says under 35. The retirement age line says 65.

Thirty years lie between them. The pothos on the desk is still there. The keyboard cover needs replacing again.