Woter AI detection.Hurry - ends Jun 29th

New Year Sales :up to 80% OFF

AI Humanize AI Translator Bypass AI AI Rewriter AI Detector

PRICING

TRY FOR FREE

Technical Thinking | AI Coding: Experience barriers are failing, tool evolution reshapes programming

Written by

Jasper Cole

Updated on:June-13th-2025

Introduction

AI is developing rapidly, and AI Coding has brought a great impact on the traditional R&D model. The author of this article summarizes the five changing trends observed, as well as his thoughts on the core skills that developers need to master in the AI era. Everyone is welcome to communicate.

1. Professional transfer and upgrading

Generative AI is indeed lowering the threshold for certain technical operations, such as automatically generating basic code snippets or assisting in completing configuration tasks that previously required specialized skills. However, this does not mean the complete demise of expertise, but rather the transfer and upgrading of expertise.

In the past, the professionalism of software development was more reflected in the "how" level, such as being proficient in writing various algorithms and accurately implementing complex logic. In the future, the focus of professionalism will shift more to "what to do" and "why to do it" (what & why). This means that the ability to accurately define problems, clearly break down complex tasks, and have critical thinking to evaluate AI-generated results will become new core professional skills. IBM's research also pointed out that although there are concerns that AI may weaken developers' basic coding skills, the more common view is that AI is enhancing rather than replacing developers, allowing them to focus more on system-level optimization and more creative work.

However, over-reliance on AI does have potential risks, which may cause developers to degenerate in basic programming skills. Therefore, this "de-professionalization" should be understood as a reshaping of professional connotations. It actually places higher demands on "professionals", requiring you to think about problems from a higher dimension and effectively master AI, a powerful tool, rather than just passively accepting its output.

2. Efficiency comes from high quality and focus

It's atmospheric and you don't have to concentrate. You can get the work done while sipping milk tea.

This "satisfaction" comes from the fact that AI frees developers from a lot of repetitive and boring "physical work". For example, AI can quickly generate boilerplate code, perform preliminary code formatting, and even assist in writing unit test frameworks. This allows you to devote more of your precious energy to more creative and challenging tasks, such as thinking deeply about system architecture, solving core complex logic, and conducting forward-looking innovative designs.

GitHub's survey data also shows that developers who use AI coding tools usually experience higher work happiness and satisfaction, and are more likely to enter a "flow" state. Therefore, this may not be a simple "low focus", but more like a "high-quality focus" - you can more effectively allocate limited focus resources to links with higher value density.

Of course, if AI is simply viewed as a tool for "slacking off" and you expect AI to do all the work while you drink milk tea, then there is no way to improve efficiency, and in the long run, this is also not conducive to personal career growth. The key is how to properly use the sense of ease brought by AI and transform it into the internal motivation for deep thinking and continuous innovation.

3. Reusability is still an iron rule

Introducing a calendar component library vs directly generating a calendar component that meets the minimum business needs.

The concern about "de-componentization" is because AI does have the ability to quickly generate "minimized components" that seem to meet current specific business needs. However, the complexity of software development is not only about the immediate realization of functions, but also about its long-term maintainability, scalability and overall stability. GitClear's research points out that AI-assisted coding may lead to an increase in the proportion of duplicate code in the code base, because developers may no longer actively refactor code for reuse as in the past, but tend to let AI generate new code for each similar scenario. This practice will quickly accumulate technical debt, causing the software system to become increasingly bloated and difficult to maintain.

The iron rule of reusability has been shaken, but it has stabilized again. Reuse is still necessary, and architectural design is still needed, because efficiency is paramount. AI is really cool, but 4 terminals x 4 times the workload and the risk of inconsistency are not cool at all.

The "iron law of reusability" has not only not been shaken in the AI era, but has become even more important because of the potential chaos that AI may bring. A well-designed reusable component is usually fully tested and verified in actual applications, and its reliability and robustness are much higher than ad hoc generated code snippets. In contrast, although AI-generated code is fast, its quality, readability, maintainability and even security may vary, requiring you to invest extra effort to identify and correct it.

Especially in the scenario of multiple terminals (such as iOS, Android, Hongmeng, Web, and mini-programs), if each terminal relies on AI to generate an independent set of so-called "minimized components" without a unified architecture design and reuse strategy, the result will almost certainly be disastrous. The inconsistency of functional performance of each terminal, the waste of resources caused by repeated development, and the sharp rise in subsequent maintenance costs will all make the initial sense of pleasure of "one shuttle" disappear.

Therefore, architectural design is still the core link of software development in the AI era. The architect's responsibilities have not been reduced, but they need to think about how to effectively integrate AI capabilities into the existing componentization and reuse system. For example, AI can be used to assist in generating component code that conforms to established architectural specifications and design patterns, or AI can be used to intelligently transform, optimize, and generate documents for existing component libraries. The principle of efficiency first has not changed, and true efficiency comes from well-thought-out architecture and high-quality reuse.

4. Full-stack developers and domain experts

One position with multiple responsibilities: front-end engineer, full-stack engineer, and full-time engineer.

When many coding, testing, and even some simple design tasks that originally required specific field skills can be completed with the efficient assistance of AI, the capabilities of individual engineers seem to have been greatly expanded. McKinsey's report also pointed out that AI will drive developers to develop more comprehensive full-stack capabilities, and even give birth to so-called "AI stack developers", that is, compound talents who can skillfully use AI tools and platforms for software development.

This sounds promising, but it also places unprecedented high demands on the comprehensive quality of engineers. Not only do you need to master a wider range of technology stacks, but you also need to have strong rapid learning capabilities, the ability to integrate cross-domain knowledge, and the ability to collaborate efficiently with AI and accurately judge and ultimately be responsible for its output results.

However, "one post with multiple responsibilities" may also bring new problems. People's energy is limited, and overly broad responsibilities may lead to a "jack of all trades, master of none", that is, having some knowledge in multiple fields but lacking sufficient depth. The team collaboration model may also need to be adjusted, and perhaps a more flexible and dynamic skill combination will emerge, rather than simply putting all responsibilities on a few "full-fledged engineers".

More importantly, deep expertise and long-term experience are still indispensable in solving complex, difficult and critical problems. AI may be able to efficiently complete 80% of general and repetitive tasks, but the remaining 20% of key problems that determine the success or failure of the system and are related to core innovation are still highly dependent on experts who have been working in specific fields for many years.

5. Shift left, shift left again

Shifting testing left: The natural fit between TDD and AI programming

TDD is in line with AI programming because humans do not need to know how AI specifically does things or thinks, as long as the final result works well.

TDD (test-driven development) is an insightful view of AI programming. In TDD mode, developers first write test cases to define the expected software behavior, and then write code to pass the test. AI can help generate code based on these clear test cases, which allows you to focus more on accurately defining the "expected behavior" (that is, writing high-quality tests) instead of getting entangled in specific implementation details. AI can even help generate some test cases to help identify some boundary scenarios or abnormal situations that developers may ignore, thereby improving test coverage and quality.

However, the "black box" nature of AI-generated code also brings new challenges to testing. How can we ensure that the internal logic of AI-generated code is correct, safe, and free of potential bias or backdoors? If we rely too much on AI to generate test cases, it may also lead to low-quality design of the test cases themselves, or fail to cover the truly critical business logic. Therefore, even in the era of AI-assisted programming, it is still crucial for humans to think carefully about test strategies, strictly review AI-generated code, and focus on testing core business logic.

Shifting R&D Left: Direct Transformation from Requirements to Code

prompt2code and prd2code are formally the same thing.
Lowcode/nocode is a successful example of R&D left shift in specific scenarios.
The essence of shifting left is to reduce intermediate links, reduce information loss and improve efficiency.

"Prompt2code" and "prd2code" (generating code directly from requirement documents) are indeed the embodiment of the concept of R&D left shift. Their core goal is to shorten the path from requirement understanding to code implementation, minimize the loss and distortion of information during transmission and conversion, and thus improve end-to-end development efficiency. The success of low-code/no-code platforms in specific business scenarios (such as internal enterprise applications, simple website construction, etc.) also confirms the effectiveness and huge potential of this "left shift" trend. The intervention of AI can theoretically convert requirement documents described in natural language into executable applications more directly and quickly.

But this also brings severe challenges. The inherent ambiguity, ambiguity and incompleteness of natural language itself are huge obstacles that "prd2code" needs to overcome. How to ensure that AI can accurately understand complex and sometimes even contradictory business requirements and generate code that fully meets expectations? This in turn places higher demands on the capabilities of demand proposers (such as product managers and business analysts) - they may need to learn how to write clearer, more precise, more structured, and even more "AI-friendly" requirements documents.

In addition, "left shift" also means redefining the boundaries of responsibility. If AI directly generates problematic code based on the requirements document, who should bear the responsibility? Is it the person who proposed the requirements, the provider of the AI tool, or the engineer who ultimately reviewed the code?

The concept of DevOps is to shift the focus and practice methods of operations to the left of the development stage, so as to achieve close collaboration between development and operations. Similarly, the "R&D shift left" caused by generative AI may mean that demand analysis, system design, and even some user experience (UX) design work need to consider the capabilities and integration methods of AI earlier, and let AI intervene in the entire creative process earlier, rather than just as an assistant in the coding stage.

6. What are the 10% of core skills?

Long-term planning ability: Humans can break down a super-large project into a series of interlocking fine-grained tasks, and then multiple human colleagues can complete the project in a few weeks to months. Everyone is an agent with super long-term planning ability, learning/reflection, and generalization ability.
Decision-making/problem-solving ability: Make correct decisions based on multi-faceted information/analytical reasoning and take responsibility (AI is not to blame for this).
Ability to collaborate with others: mobilize resources and handle interpersonal issues.

Kent Beck once said: "The value of 90% of my skills just dropped to $0. The leverage for the remaining 10% went up 1000x. I need to recalibrate."

This passage accurately summarizes the dramatic differentiation of skill value in the AI era. The "90% of skills whose value has been reduced to zero" he mentioned are likely to be those repetitive, fixed-pattern coding implementation, basic debugging, routine information retrieval and other tasks that can be efficiently replaced by AI. These were the "crafts" that engineers relied on to survive in the past, but in the face of the powerful capabilities of AI, their scarcity and uniqueness have indeed been greatly reduced. This does not mean that these skills are completely useless, but that they are no longer the key to determining the core value of engineers.

So, what exactly are the "10% skills that leverage 1,000 times"? These are the advanced cognitive abilities and soft skills that are unique to humans and difficult for AI to achieve. People with these skills can effectively harness AI and use it as an extension and amplifier of their own capabilities, thereby creating tremendous value far beyond the past. Specifically, these key skills include:

1. Long-term planning and complex problem definition capabilities: AI can help solve the problems you clearly raise, but it cannot discover and define complex problems that are truly commercially valuable and socially significant. Transforming vague business visions into clear technical strategies and executable roadmaps, designing robust and scalable complex system architectures, conducting long-term strategic planning, and foreseeing potential risks and dependencies requires strong abstract thinking, systematic thinking, and rich experience. You can break down a super-large project into fine-grained tasks that AI can understand and gradually execute, and foresee the challenges and coordination issues that may arise when integrating these AI "parts", which is the starting point of "1000x leverage".

2. Critical thinking, decision-making and responsibility: Content generated by AI, whether it is code, analysis reports or design plans, may contain errors, biases, omissions or limitations that do not conform to specific scenarios. Being able to use critical thinking to examine AI output, discern its authenticity and quality, combine multiple information (including business intuition, industry insights, ethical considerations and value judgments that AI cannot provide) to make key decisions, and take ultimate responsibility for the consequences of these decisions - this is the core embodiment of human wisdom and responsibility. AI will not "take the blame" for project failures or wrong decisions, but people with these 10% skills will, and that is why their judgment and decision-making become invaluable.

3. Deep insight and creative problem-solving capabilities: AI is currently better at finding the best solution or pattern matching in existing knowledge and data patterns, but it often seems powerless for disruptive innovations and original solutions that require breaking out of the inherent framework and conventional thinking. The ability to see into the essence of the problem, discover hidden connections, propose new ideas, and design clever solutions that AI "cannot think of" - this creativity derived from human inspiration, intuition and imagination is difficult for AI to replicate. When AI can help deal with the 90% of tedious work, you will have more energy to carry out this kind of deep and creative thinking and exploration.

4. Collaboration, communication and leadership: Software development is essentially a social activity that relies heavily on teamwork. Effectively communicate across teams, clearly express complex ideas, coordinate resources from all parties, motivate team members to realize their potential, handle complex interpersonal relationships and potential conflicts, and build and maintain trust - these abilities to deal with people have not only not been weakened in the AI era, but have become more important due to the complexity of the project and the increased requirements for interdisciplinary collaboration. You need to lead a diverse team that may include human colleagues and AI "assistants" to move towards a common goal. This kind of leadership and influence is an important part of achieving "1000x leverage".

5. The ability to continuously learn, quickly adapt and "recalibrate": As Kent Beck emphasized, "I need to recalibrate." Technology is developing at an astonishing speed, and the capabilities of AI are constantly expanding. Being able to maintain an open mind, quickly learn new knowledge, master new tools, adapt to new methods, and actively adjust one's own skill structure and working methods, this metacognitive ability and lifelong learning growth mentality are the fundamental prerequisites for maintaining and developing the critical 10% of core competitiveness.

Kent Beck's feelings reflect very truly that AI is forcing the entire industry to conduct a profound "revaluation of value". Those who can accurately identify and fully invest in developing and applying these "10%" core skills will truly become leaders and value creators in the AI era, and their influence will be magnified hundreds or thousands of times by AI, a powerful "lever". If you only stay at the level of those "90%" skills that can be easily replaced by AI, then the sense of crisis in your career development will be very real and urgent.

7. The experience of “old hands” is also important in the AI era

AI has narrowed the experience gap in some aspects, especially for problems that can be solved by learning existing patterns and information retrieval. Studies have also shown that junior developers can usually get greater immediate productivity gains from AI coding assistants. Newcomers can use AI to quickly get started and complete tasks that would have required a certain amount of experience in the past, which undoubtedly lowers the entry barrier.

But does this mean that the deep experience accumulated by the "old masters" is worthless? I don't think so.

First of all, the experience of "old masters" is not just the ability to solve known problems, but more importantly, their ability to deal with unknown, complex and ambiguous problems, as well as their judgment based on intuition and experience at critical moments. These deep experiences and wisdom are difficult for AI to simply learn and replicate. AI may be able to imitate specific problems that "old masters" have solved, but when faced with a new and unprecedented challenge, the experience of "old masters" can often guide the direction.

Secondly, AI tools themselves also require experience to master. Experienced developers can usually ask AI questions more effectively (i.e. write higher quality prompts), more accurately judge the pros and cons and potential risks of AI-generated results, and can more effectively correct, optimize and integrate them. They know when to trust AI, when to question AI, and when to intervene manually for in-depth adjustments. As Addy Osmani pointed out, AI coding tools tend to be more beneficial to experienced developers because they can constantly evaluate, correct and improve AI outputs; junior developers who lack critical thinking may be more likely to blindly accept AI results, and may write code that appears to work but is actually fragile or has hidden dangers.

Furthermore, the value of "old hands" lies in the fact that they usually have a deeper understanding of the business field and a more comprehensive understanding of the system, which are difficult for AI to fully grasp through code learning alone.

Therefore, rather than saying that "random punches kill the master", it is better to say that AI is forcing the "master" to change the way they embody their core values. They need to change from being the executors who used to do things themselves to being more strategic planners, complex system designers, key decision makers, quality controllers, and knowledge transmitters and leaders of the team. "Newborn calves" need to be vigilant and not ignore the cultivation of basic abilities, the exercise of critical thinking, and the in-depth understanding of the nature of the problem just because they have AI, a powerful tool. Otherwise, they may always remain at the stage of "beginners who can use AI" and find it difficult to cope with more complex and challenging work in the future.

Summarize

The changes brought about by AI Coding are profound and multifaceted. It not only brings unprecedented efficiency improvement and innovation possibilities, but also puts forward new requirements for the ability model of practitioners, the way teams collaborate, and the organizational structure and value chain of the entire software industry. This is indeed an era full of challenges and opportunities. The key lies in how you understand, adapt to and actively control this irreversible wave of technology.

Technical Thinking | AI Coding: Experience barriers are failing, tool evolution reshapes programming

Shifting R&D Left: Direct Transformation from Requirements to Code

prompt2code and prd2code are formally the same thing.