What Is Happening in the Software Industry Now?
In recent years, with the emergence of general-purpose artificial intelligence such as ChatGPT and Claude Code, often referred to as AGI, companies have increasingly begun introducing AI into their workflows.
Starting in June 2026, GitHub Copilot AI made a major change. It moved away from the PRU model, or premium request units, which was based on request counts, and shifted toward a model calculated using AI Credits. As a result, the execution cost for an engineer developing a feature, reviewing PRs, organizing documentation, and performing other tasks has gradually increased.
Some developers even used up an entire month’s allowance on the very first day GitHub Copilot AI adopted the AI Credit model. This shift toward token-based pricing is becoming increasingly mainstream. It also places more restrictions on using AI within workflows. If you are not careful, your usage can run out quickly, and costs can continue to rise. Worse still, this type of pricing model is likely to become a broader trend among AI service providers.
Jensen Huang once said:
If that $500,000 engineer did not consume at least $250,000 worth of tokens, I am going to be deeply alarmed.
The question today is no longer whether we should use AI. After all, not every company is like NVIDIA, with the freedom to consume tokens at scale. The real issue is the conflict between problem-solving, money, and time.
Solving problems and controlling costs have already become conflicting forces. The future direction of the software market will be about solving problems effectively while keeping costs under control. It will not be about how many tokens you used to solve something, but about achieving the greatest benefit with the lowest possible cost.
What Should Business Managers and Development Leads Do Next?
Clearly, the first step is to identify what qualities and capabilities future technical talent should have:
- The ability to break down large problems and requirements, extract the key points, group them properly, and find the best solution
- The ability to simplify AI behavior so it can be executed at the lowest possible cost
- The ability to achieve maximum benefit with the fewest tokens
Here is an evaluation framework for companies, including both positive and negative evaluation points.
Positive Evaluation Points
If the company is willing to spend money to solve problems:
- Has this problem been troubling the company for a long time? Was it difficult to execute through conventional methods because of staffing, scheduling, or other reasons?
- Has AI effectively replaced existing talent resources in the way an expert system might? While ensuring correctness, has it saved human effort?
- Was the problem solved quickly? Did it create additional time costs?
Negative Evaluation Points
If the company is relatively conservative:
- Will this adjustment reduce future changes, or will it increase them? How should we limit the damage?
- If problems occur, are the losses acceptable? For example: time, reputation, and financial cost.
- How much benefit can this adjustment bring? Compared with problem-solving costs and personnel costs, how much money does it actually save?
- Will future maintenance costs decrease? Or will AI make them higher and unnecessarily so?
In practice, what business managers need even more is a cost plan:
- How much cost are we willing to invest in solving the problem?
- Is the risk of cost increase controllable?
- Are there audit points in place to control costs?
Regarding the common online discussion that AI will replace software engineers, my view is relatively optimistic. There are many detailed technical issues that bosses or non-engineering project managers simply cannot fully grasp. When a company truly lays off all of its software talent, that is when the real disaster begins.
That said, it is undeniable that Junior Engineers will become increasingly rare in the future. The technical threshold will become higher, because using AI well requires the corresponding knowledge. In contrast, Senior Engineers who can combine experience and expertise will become one of the strongest resources in the market.
The knowledge and experience of software professionals will become an important dividing line.
Here is a simple example. Suppose a project developed through Vibe Coding suddenly encounters a serious bug in production. Every minute the problem exists causes the company to lose money.
If you were the manager, would you want the developer to search for the answer as if rolling dice? Or would you want the developer to quickly identify the root cause?
I also like this saying:
People can never imagine what lies outside their own understanding.
The knowledge of a project member is the key to future product maintenance.
Felix Rieseberg, Engineering Lead at Claude Cowork, also said:
The biggest barrier to AI adoption is people not realizing they can ask AI to solve almost any problem… The gap isn't technical; it's psychological.
Felix Rieseberg points out that the reason people fail to use AI well lies in imagination. What you can think of determines what you can ask. When using today’s top-tier models, the core problem is not always that the result falls short of expectations. More often, it is that the user has not described the requirements clearly enough, causing the AI to produce results that do not match what was expected.
Of course, the input and output limitations of model context, as well as the model’s byte-processing issues, must first be addressed.
Next Article: Why Does AI Get Lazy?
The next article will discuss a common phenomenon:
Why is it that when two teams use the same AI, one team significantly improves its efficiency, while the other sees almost no improvement in productivity?
Many times, it is not that AI does not understand. Rather, it is...
Stay tuned.
The original article was written in Traditional Chinese, and the Japanese and English versions were translated using ChatGPT.
Top comments (0)