From last year until now, my cumulative token spending on Vibe Coding has exceeded 2,300 yuan, which works out to a monthly average of about 281.88 yuan. Data is as of September 16, 2026. Day to day I mainly use Kimi Code and Deepseek as my two model service providers, relying on AI assistance to complete a large volume of project development and experimentation.
Rough estimate: that 2,000-plus yuan investment ultimately produced thirty or forty applications of various sizes. Over a dozen have been uploaded to my blog for public download, some open-sourced, the rest kept for personal use. On average, each completed application cost only a few dozen yuan. Purely from an input-output ledger perspective, I think this spending was well worth it.
But objectively speaking, this is also the single highest expenditure I’ve made in the software-services domain over many years. What’s more unusual: every project was for personal use — no monetization revenue was generated at any point. Setting aside the surface-level numbers, I believe the deeper value of this investment goes far beyond collecting a pile of self-use tools.
This year of continuously using AI to write code, tinker with projects, and develop applications — I’ve accumulated substantial hands-on experience and gained long-term hidden returns. The most tangible change: I can now develop my own tools, progressively cutting various long-standing paid subscriptions and permanently eliminating recurring expenses.
Over the past many years, I’d been subscribing to all kinds of paid tools: 1Password, which I used continuously for over a decade, originally costing 70–80 yuan per year, raised to 90-plus yuan this year — even splitting a family plan is a persistent cost; Paste clipboard manager at about 80 yuan per year. Now both have been completely replaced by self-built tools, and those fees are gone.
On the entertainment side, I set up a multi-device music service compatible with phone, computer, and car head unit, deployed on my home mini-PC — completely replacing commercial music memberships. The library isn’t massive, but it covers everything I listen to regularly, ad-free, no subscription fees, saving over a hundred yuan per year in music-service spending.
Savings in the cloud-services domain are even more substantial. I used Alibaba Cloud’s servers, object storage, CDN and other resources for years; from January to August 2026 alone, total Alibaba Cloud spending reached 644.4 yuan. I’ve now decided to completely cut all Alibaba Cloud budgets and replace everything with self-built solutions.
This decision isn’t just about saving money — it’s a pointed boycott. Alibaba Cloud, as an industry-leading giant, blatantly practices workplace age discrimination, outright rejecting job seekers over 35. In an era where the 35-year-old workplace predicament is already widespread, a company with a market cap in the hundreds of billions openly institutionalizing and normalizing age discrimination — I consider this extremely non-compliant and devoid of vision. Given that, I have no reason to continue paying them or funding them.
I’ll be posting a list on my blog detailing the companies and products I firmly boycott, along with my reasons for each — using my own consumption choices to push back against unreasonable industry practices like workplace age discrimination.
I used to frequently sign up for training courses to learn new technologies, but looking back, all that stuff is outdated now. From here on, no training course is worth paying for. Because AI is the best and most efficient teacher. Compared to paying for in-person programming bootcamps, this “learn by doing with AI” model is far more cost-effective and practical for me. Over this year, leveraging AI development as a catalyst, I tinkered with software and hardware across the board: switched my main computer to Linux, repeatedly flashed and debugged my phone, developed all kinds of work scripts and utility tools entirely with AI assistance. I also completed personalized AI customization for my car head unit and smartwatch, making all my smart devices’ interaction logic and feature adaptation perfectly match my own habits.
But after a full year of deep practice, I’ve also completely shattered the blind filter around “AI boosting efficiency.” I’ve always felt that the AI-omnipotence hype on the market is, at its core, a false proposition.
Much of the time I’m like a primitive onlooker, watching AI run at high speed, burning tokens, rapidly producing runnable code and applications. Seeing the results does produce a real sense of accomplishment, but the hardcore technical capabilities and engineering thinking that settle in haven’t undergone any qualitative improvement. And what appears to be a money-saving alternative is essentially cost-shifting: before, you paid commercial vendors for finished services; now you pay AI model providers, while simultaneously investing large amounts of personal time and energy manually sorting logic, debugging code, fixing bugs. Factor in all these hidden costs, and the total is actually higher.
AI writes code extremely fast, but it produces bugs at an equally astonishing rate. Code that runs and code that’s stable enough for commercial production use are completely different things. I’ve come to deeply realize: people without engineering capability and product thinking who use AI for development are just piling crap on top of crap. The vast majority of people on the market who ride the AI hype wave and blindly depend on AI development mostly lack foundational technical depth and engineering implementation ability. Things written by stacking prompts are fine for personal play, but absolutely insufficient for commercial use — there are far more issues to consider.
AI has never been omnipotent. Many real-world problems it simply cannot solve independently, and it frequently burns tokens ineffectively, gets stuck in loops, or cycles through debugging with zero progress. At its current stage, AI always requires deep human involvement and precise oversight. The user’s own technical depth and ability to wield it remain the ultimate core moat.
It’s absolutely not as simple as writing a few prompts or configuring some Skills constraints to fully control AI. Even when you pre-constrain permissions in instructions and prohibit violations, AI still oversteps. To mitigate risk and enforce proper use, you must rely on programmatic hard constraints — for example, using KMS to manage credential permissions, restricting AI from directly reading core keys, only opening application-level invocation rights, eliminating vulnerabilities at the foundation.
For complete beginners, getting AI to run a simple demo might not be hard, but achieving engineering-grade implementation and commercial-grade productization without professional technical moats and implementation experience is essentially impossible. I’m extremely averse to the brainwashing rhetoric on the market about “one-click AI entrepreneurship” and “solo company, easy landing.” In the first half of the year I personally attempted independent development and self-directed project implementation, hit every possible pitfall, and only then understood that the difficulty far exceeds imagination.
AI can only replace the single step of “writing code.” Business communication, channel development, project operations, risk management — these core functions AI is completely powerless to handle. And AI itself is highly unpredictable, frequently producing elementary errors. Just recently, I asked AI to calculate a Unix millisecond timestamp — it not only produced a wrong result but claimed it had “mentally calculated” it, which was laughable. Meanwhile, the quality of open-source AI tools varies wildly — Claude Codex, OpenCode, etc. — opencode appears open-source and free but is riddled with bugs; even simple terminal copy-paste can’t be reliably implemented. This taught me that model selection and tool vetting are critically important steps in AI development.
This year’s 2,000-to-3,000-plus yuan AI investment is essentially equivalent to self-funding an “AI hands-on training course.” Learning barriers lowered, practical efficiency improved — but that’s about it. This capability won’t help me get promoted or earn a raise, and it won’t broaden my career development path. At my current age, even finding a basic job paying just over 10,000 yuan per month is extremely difficult. The so-called AI efficiency boost has no substantive career value for an ordinary worker or mid-career professional like me.
What’s even more ironic: AI improved efficiency on individual problems, but caused the total volume and complexity of problems across the entire workflow to surge dramatically. Before, troubleshooting a problem meant a simple reference lookup to identify the root cause; now that development universally uses AI to write code, bug counts have exploded, logic is chaotic, documented rules and actual runtime behavior are completely misaligned, and the difficulty of problem investigation and fault localization has doubled.
Every industry is immersed in the AI efficiency carnival. Companies would rather pour massive funds into building AI infrastructure and chasing the AI track than raise employee salaries or improve benefits. On the contrary, AI’s proliferation has further intensified workplace exploitation: on top of unchanged existing workload, companies require employees to continuously learn AI tools, adapt to AI workflows, and keep increasing their load. After adopting AI, people haven’t become more relaxed — they’ve become more exhausted. What I know is that some company managers, before AI, might have provided only rough, ballpark data. Because development costs were too high, there was no way to achieve such fine-grained management. But with AI, they can use it to fill that gap, making the entire management more granular. For example, before they just roughly estimated some numbers and that was enough to satisfy the boss; now with AI they do more precise calculations, the boss’s expectations rise — like tracking each person’s working hours and efficiency, further instrumentalizing people, the same system as food-delivery riders.
Moreover, large numbers of formerly permanent full-time positions have been replaced with outsourced roles, with the actual work content unchanged. Take my current job: busy from morning to night, take-home pay barely reaching 10,000 yuan after deductions — half of what I earned before. And as an outsourced employee, my rights protection, salary, and benefits are completely different from permanent staff; the moment they don’t need me, I get kicked out without even needing to discuss compensation. I’ve come to see that every technological revolution ultimately has its dividends captured by big capital and enterprises, leaving ordinary people at the bottom with only more severe exploitation, a more hyper-competitive workplace environment, and harsher survival pressure. This is simply not a win-win game. This round of AI will make employment positions increasingly scarce, push the vast majority into unemployment, and then everyone will be fighting tooth and nail for the few remaining positions. It’ll become increasingly competitive — work that previously required only a vocational diploma now demands a bachelor’s or master’s degree. This is already happening, and I’m living through it. Many people simply cannot find employment at all. So what meaning does this technology hold for human development, for human happiness? At least I haven’t felt any dividend AI has brought me. Instead it left me unemployed for half a year, then took a pay cut of more than half. You might say AI’s dividend to me was writing a few apps that saved me some money. I don’t count that. As I said above, the actual cost is higher. Building your own applications means buying servers, spending time writing APIs, building front-end and back-end. Add that time and token costs together and it’s more expensive than just buying software. Unless you have channels to sell the app as a service and amortize the cost — but the vast majority of people don’t have the ability to sell products.
The absurdity of this era lands with particular reality on every ordinary family. My wife also lost her job at the end of August. She worked diligently at a company for nearly ten years, holds a 211-university degree — from an academic standpoint far more impressive than my vocational-college liberal-arts background — yet in this era she ultimately left in quiet defeat, with the company still owing her a full year’s unpaid salary. Their company also chased the AI/GPU wind and lost everything, couldn’t even pay employee wages. She doesn’t want to look for work anymore; she feels employment is pointless. I actually feel the same way. But I’m carrying a mortgage, a family, and a child — I can’t afford to stay unemployed indefinitely.
Looking back at these past few years: technology iterating rapidly, industries constantly transforming — yet none of it has made ordinary small families and ordinary people better off. All the survival costs and workplace pressure ultimately cascade down layer by layer onto ordinary people. Under the banner of “debt resolution,” the debt ends up on residents’ heads. Responsibility and risk all land on ordinary people. You say many companies use outsourcing now — isn’t that just shifting risk and responsibility? On one hand they shift all risk and responsibility outward; on the other hand they demand employees show responsibility and commitment. This asymmetric demand for full accountability is the most outrageous and disgusting thing.
AI is a tool, a trend — but it’s not ordinary people’s salvation. It can optimize efficiency, save some living costs, but it can’t change the predicament of rank-and-file workers, much less dissolve the anxiety and heavy pressure this era leaves for ordinary people.
There’s something else I want to say: in the months ahead, more and more people may become more selfish and private. Before, people would share things online — code and so forth. But increasingly people may realize that after sharing something online, it effectively becomes fuel for model companies — we’re making wedding dresses for others while we ourselves can’t benefit.
评论
0 条评论