The AI Coding Hangover: When Token Bills Outpace Payroll Savings

I'm not sure how much AI has already covered in everyone's daily life, but for programmers, AI programming is now simply too good to use.

The speed and quality of code output can only be described as "changing with each passing day". I currently see that in all project teams and all development work across the company, almost all the code produced carries the "imprint" of AI.

There's no way around it; AI is too fast, and most importantly, it no longer requires people to painstakingly think about implementation logic.

All you need to do is "translate" requirements into a direction that AI can understand and execute accurately, then hit "Enter", and your work is done.

But does AI really only bring convenience?

And without any "side effects" at all?

Impossible, that wouldn't conform to the "conservation of energy", right? Haha!

---

Let me show you a few more news items: Tencent and Alibaba vigorously promoted AI programming internally, directly issuing Tokens to encourage employees to switch to AI.

Just a few months later, they issued another internal notice, suspending free Token usage.

They abolished the previous uniform free Token quota for all employees and the "Token consumption ranking" assessment mechanism.

Employee Token quotas are no longer fixed but dynamically allocated based on job requirements.

In January, Microsoft issued an internal memo encouraging the use of Claude Code for development, urging engineers to reshape their workflows with AI.

In June, another memo was issued, fully switching to Copilot and cutting off the supply of Claude Code.

Uber's CTO excitedly announced internally that 95% of their engineers are now using AI tools, 84% have entered the era of agent coding, and 70% of code comes from AI submissions.

Then the entire AI budget for 2026 (about $3.4 billion) was burned through in four months.

Rumors online say miHoYo ran up a Token bill of 2 million yuan in a single night.

Meta "burned" 60 trillion Tokens in 30 days.

......

These examples are not meant to say that AI is bad, that AI is unimportant, or that AI is worse than humans.

It's about the trade-off between cost and risk!

The primary purpose for inventing AI tools was actually to "reduce costs and increase efficiency", with the core goal of reducing the highest proportion of company costs, "human resource costs".

For bosses, the perfect solution is 2 people + AI doing the work of 5 people and getting the salary of 3 people.

With the assistance of AI, many business requirements no longer follow the original scheduling rules; a task you estimated would take 1 month might only take 1 week with AI.

But! But! But!

Is that really the case? Obviously not!

First, 2 people will probably still do the work of 2 people. If you ask them to do more, they'll ask for a raise, or they'll just "slack off", and you can't do anything about it.

It's not like firing people means the remaining ones will increase efficiency; there's never been such a rule throughout history.

Second, the cost of AI far exceeds your estimates. Logically, the monthly cost of AI shouldn't exceed an employee's monthly salary, but when the complexity of requirements and the required context length increase, it doesn't necessarily follow this "script".

Most critically, AI's capability has boundaries, the context length is limited, and your Tokens are also limited.

When you hand all your work over to AI, if something goes wrong, AI cannot "take the fall" for you. Are you going to ask Doubao or Claude Code to apologize?

Once a problem occurs, "reducing costs and increasing efficiency" becomes "reducing costs and increasing laughs".

If your client doesn't know this is AI output, it's fine, but once they find out, the consequences...

---

So, do you lay people off or not?

Lay them off! Then what if a problem occurs? Can you bear that loss?

Don't lay them off! Employee costs remain unchanged, and AI costs are added on top. Not only did you not "reduce costs", you actually "increased costs".

Most importantly, all domestic Coding Plan prices have now gone up!

All of them!

Not just one or two!

Employee costs may have been reduced, but AI has instead become the largest cost source.

At this point, it's a bit awkward. If you switch back to humans, you can't accept that level of efficiency.

If you keep using AI, you have to accept the continuously rising costs.

And let's make a bold prediction: we are still in a "blue ocean" market phase; the pie isn't big enough yet.

Once it enters the "stock" phase, the cost of AI will inevitably rise further. When everyone raises prices then, how will you, sir, respond?

Perhaps there is never absolute "cost reduction and efficiency increase". Using any tool can only guarantee pushing efficiency a tiny bit higher, just that tiny bit.