What AI still can’t replace in software development

Junior developers believe AI improves learning, but senior developers question whether it builds real understanding

Artificial intelligence has changed how new developers learn software engineering. For many entering the profession today, AI is not an optional tool. It has been part of the learning process from the beginning. That changes how they experience software development and how they judge their own progress.

The Q2 2026 Dev Barometer highlights a striking difference in perception. Eighty-five percent of junior developers say AI has improved their understanding of software development. At the same time, only 16% of senior developers believe juniors fully understand the AI-generated code they submit. On the surface, these findings seem contradictory. They are not.

The important point is that junior developers are not ignoring the risks created by AI. The survey shows that 50.5% of juniors believe AI is making the junior developer role less relevant. Senior developers reach almost the same conclusion, with 53.3% agreeing. That tells us something important. Juniors understand the market is changing. Their optimism about AI does not come from believing they are protected from disruption.

The real issue is how people define understanding. A developer who learned programming before AI became common usually built knowledge by solving problems manually, making mistakes, and debugging without assistance. A developer who learned with AI available from day one experiences a very different process. AI becomes part of how learning happens rather than a separate productivity tool.

For business leaders, this distinction matters because perceived capability and actual capability are not always the same. AI can produce working code very quickly. That creates visible output, but output alone does not guarantee deep technical understanding. The difference becomes clear when systems become more complex, when AI produces incorrect recommendations, or when architectural decisions require judgment instead of code generation.

This does not mean organizations should reduce AI adoption. The opposite is true. AI will become an essential capability across engineering organizations. The competitive advantage comes from ensuring that developers understand why the generated solution works. Companies that combine AI adoption with strong technical development programs will likely build engineering teams that remain effective as AI continues to improve.

For executives, this changes how technical talent should be evaluated. Measuring productivity alone is becoming less useful because AI increasingly raises everyone’s output. The stronger signal is the ability to explain decisions, evaluate AI-generated solutions, identify hidden risks, and solve unfamiliar problems independently.

AI can accelerate learning, but it can also hide knowledge gaps that become expensive later

One of the biggest questions facing technical education is not whether AI helps people learn. It clearly does. The harder question is whether AI changes how people develop judgment.

Learning has always depended on more than finding the right answer. It depends on recognizing why an answer is right, understanding why alternatives fail, and discovering gaps in your own thinking. AI changes that process because it often removes the friction that traditionally forced people to think through problems step by step.

John Flavell, widely recognized for introducing the concept of metacognition, argued in 1979 that people improve their understanding by monitoring their own thinking. That process depends on recognizing mistakes, uncertainty, and incomplete knowledge. If AI consistently removes those experiences, developers may become less effective at identifying what they do not understand.

This does not mean AI weakens learning by definition. The outcome depends on how it is used. There is a major difference between asking AI to automate repetitive work after understanding a concept and asking AI to generate solutions before building that understanding. In the first case, AI increases productivity. In the second, it can reduce opportunities to develop the mental models that support long-term decision-making.

The survey reflects this distinction. Twenty-four percent of junior developers said writing code from scratch without AI is the task they feel least confident performing. Yet only 5% believe that ability is a critical hiring skill today. From today’s hiring perspective, that conclusion makes sense. Most software organizations now expect developers to use AI. However, hiring requirements and long-term capability are not always the same thing.

As developers advance into senior engineering, architecture, or technical leadership, the job shifts away from writing individual functions and toward evaluating trade-offs, identifying risks, designing systems, and making decisions under uncertainty. Those responsibilities depend heavily on conceptual understanding rather than speed alone.

For executives, this creates an important management challenge. AI adoption should not focus exclusively on efficiency metrics such as coding speed or feature delivery. Organizations should also measure whether engineers can explain system behavior, review AI-generated code critically, and make sound technical decisions when AI provides incomplete or incorrect answers. Those capabilities become increasingly valuable as AI handles a larger share of routine development work.

The goal is to ensure AI expands human capability instead of replacing the learning process that creates expertise.

Critical thinking remains the most valuable skill

One of the strongest findings in the Q2 2026 Dev Barometer is not about AI itself. It is about what both junior and senior developers believe will continue to matter as AI becomes more capable.

Across multiple survey questions, respondents consistently ranked critical thinking, analytical reasoning, problem-solving, and systems-level understanding above AI-specific skills. This consistency is significant because the same conclusion appeared in structured questions, ranking exercises, and open-ended responses. When different survey formats produce similar results, confidence in the finding increases.

Junior developers recognize this clearly. When asked which skill matters most for getting hired today, 48% selected analytical thinking and problem-solving. Only 18% chose proficiency with AI tools. This suggests that even developers who have learned alongside AI understand that employers are looking for more than the ability to write effective prompts.

Senior developers reinforce the same message from another perspective. The skills they most frequently find missing in junior engineers are understanding how systems work from end to end and the ability to break complex problems into smaller, manageable parts. These are not AI skills. They are decision-making skills that become increasingly valuable as software systems grow in scale and complexity.

The survey also asked respondents to evaluate which competencies are most difficult for AI to replace over the next three years. Critical thinking and analytical reasoning received the highest average score, ahead of security expertise, adaptability, and prompt engineering. That ranking reflects an important shift in how organizations should think about workforce development. AI capabilities will continue improving, but evaluating information, identifying weak assumptions, and making sound decisions remain fundamentally human responsibilities.

The open-ended responses strengthen this conclusion. Without predefined answer choices, respondents naturally emphasized critical thinking, foundational knowledge, and problem-solving far more often than AI literacy. This suggests these priorities are not being influenced by survey design. They represent what experienced developers genuinely believe creates long-term value.

For executive teams, this has direct implications for hiring, workforce planning, and capability development. AI proficiency should absolutely be encouraged. However, AI skills have a shorter competitive lifespan because the underlying tools evolve rapidly. Strong reasoning, structured problem-solving, and systems thinking remain valuable regardless of which AI platform becomes dominant.

Organizations that invest only in AI tool adoption risk creating teams that become dependent on the current generation of technology. Organizations that invest in durable thinking skills alongside AI adoption are more likely to build technical teams that adapt successfully as both technology and business requirements continue to evolve.

Senior developers differ on AI’s role in learning

The survey’s open-ended responses reveal a more detailed discussion than multiple-choice questions can capture. While senior developers broadly agree that critical thinking matters, they do not all agree on how developers should build it in an AI-driven environment.

The analysis identified four distinct viewpoints. The largest group argued that AI is simply the environment developers now work in, while judgment is the real competitive advantage. This perspective emphasizes knowing when to trust AI, when to question its output, and when to reject it completely. Technical knowledge remains important, but the ability to evaluate information becomes the defining skill.

A second group placed technical fundamentals at the center of learning. These respondents argued that developers should first understand programming logic, algorithms, databases, computer architecture, and the underlying principles of modern AI systems before depending heavily on AI-generated solutions. Their concern is not AI itself. Their concern is building enough expertise to recognize when AI is producing poor recommendations.

A third perspective focused on problem decomposition. According to these respondents, the ability to divide complex challenges into smaller, logical components is the core capability that allows developers to work effectively with AI. Better questions generally produce better AI outputs, and developing those questions depends on structured thinking rather than tool proficiency.

The fourth group viewed AI literacy as the primary capability for future developers. While this perspective recognizes the importance of critical thinking, it argues that learning to collaborate effectively with AI should become a central professional skill rather than a complementary one.

What is particularly notable is that the first three perspectives together account for the majority of responses. Although they differ in emphasis, they all begin with the same assumption: durable technical understanding supports good judgment. AI then becomes a force multiplier rather than a substitute for expertise.

One anonymous senior developer summarized this view clearly: “The developers who will grow are the ones who use AI to accelerate what they already understand, not to replace the understanding itself. Learn the concepts first, then let AI help you move faster.”

This distinction has practical implications for business leaders. Organizations should not frame AI adoption as replacing traditional technical development. Instead, AI should become part of a broader capability strategy that strengthens conceptual understanding, independent reasoning, and technical judgment. These qualities determine whether employees can continue making sound decisions as AI systems become increasingly capable.

Both Pellegrino and Hilton’s work on transferable competencies and the revised Bloom’s Taxonomy emphasize that higher-order skills, including analyzing, evaluating, and creating, remain durable across technological change. AI changes how work is performed, but it does not eliminate the need for these capabilities.

For executives making long-term investment decisions, this is an important distinction. Technology platforms will continue changing. Durable cognitive skills remain valuable regardless of which tools dominate the market in the future.

The biggest challenge for technical education is how quickly learning can adapt

Generative AI has changed software development faster than most education systems can respond. This is not unique to universities, coding bootcamps, or corporate training programs. Every learning model faces the same challenge: technology now evolves on a much shorter cycle than curriculum development.

The Q2 2026 Dev Barometer shows an interesting difference in perception. More than one-third of senior developers believe formal education does not adequately prepare people for the AI era. In contrast, over 70% of junior developers say they feel at least somewhat prepared by their training.

This difference should not immediately be interpreted as a failure of education. It reflects the different perspectives of the two groups. Junior developers often evaluate preparation based on whether they can enter the workforce. Senior developers evaluate it based on whether new hires can perform effectively in increasingly complex production environments over time.

Curriculum lag is the central issue. Educational institutions require time to redesign courses, validate content, train instructors, and implement changes. AI capabilities, however, can shift meaningfully within months. No education system, public or private, academic or commercial, is currently designed to move at that pace.

Importantly, the survey suggests that the underlying technical foundations remain relevant. The largest skill gaps identified by both junior and senior developers are not new disciplines created by AI. They are situations where existing knowledge must be applied under real-world conditions.

Working with large legacy codebases, designing systems that operate at scale, and critically evaluating AI-generated code all require applying established software engineering principles in environments that contain uncertainty, complexity, and competing priorities. These are difficult experiences to recreate fully inside a classroom.

This distinction matters for business leaders because it shifts the conversation away from replacing existing technical education. Organizations do not necessarily need graduates who have studied completely different material. They need graduates who have had more opportunities to apply what they know in realistic environments.

That has implications for corporate learning strategies as well. Internal development programs should focus less on introducing entirely new technical concepts and more on creating structured exposure to production systems, code reviews, architecture discussions, operational incidents, and collaborative engineering practices. These experiences accelerate the transition from theoretical knowledge to professional judgment.

Research by Jean Lave and Etienne Wenger, who argued that professional knowledge develops through participation in real communities of practice. Experience matters because it exposes people to decisions that have consequences, competing priorities, and imperfect information. Those conditions are difficult to simulate but define modern software engineering.

For executives responsible for workforce planning, this means that educational partnerships should extend beyond curriculum design. Companies that create opportunities for students and early-career developers to engage with real engineering work can strengthen both recruitment and long-term capability development.

Practical experience has become one of the strongest indicators of developer readiness

The survey sends a clear message from both sides of the hiring process. Classroom learning remains important, but employers increasingly expect candidates to demonstrate that they can apply their knowledge in real development environments.

Junior developers recognize this need. When asked what their education should emphasize more, nearly half selected real-world project experience. Another 20% requested additional internships and hands-on work. Together, these responses show that almost 70% of juniors want more opportunities to learn through practical application rather than additional theoretical content.

Senior developers reached almost exactly the same conclusion from the employer’s perspective. When evaluating whether a junior developer is ready for professional work, they placed the greatest value on real project experience, internships, and performance during practical coding exercises.

This alignment is significant because it suggests the market is not rejecting formal education. Instead, employers are looking for stronger evidence that academic knowledge can be applied effectively in production environments.

For organizations, this changes how talent pipelines should be designed. Partnerships with universities, apprenticeship programs, internships, and project-based learning are no longer simply recruitment channels. They are strategic investments that shorten the distance between education and productive employment.

This is especially important as AI automates more routine development tasks. Entry-level engineers will increasingly be expected to contribute to larger systems, collaborate across teams, review AI-generated outputs, and understand the operational impact of technical decisions earlier in their careers. These capabilities are difficult to develop through lectures alone.

Companies can strengthen this transition by integrating experienced engineers into learning programs through mentoring, structured code reviews, pair programming, and supervised work on production systems. These experiences expose junior developers to decision-making processes that AI cannot fully teach.

Educational institutions also have an opportunity to evolve. Rather than separating theoretical instruction and practical experience into different stages of learning, programs can integrate both throughout the curriculum. Students gain a deeper understanding when concepts are reinforced through continuous application instead of waiting until graduation to experience professional software development.

For executive leaders, this is ultimately about reducing execution risk. Developers who have worked on real projects generally adapt more quickly, require less onboarding support, and are better prepared to contribute to complex engineering teams. As AI raises baseline productivity, demonstrated experience applying technical judgment becomes an increasingly valuable hiring signal.

The greatest long-term risk is AI producing future leaders without deep technical foundations

Much of the public discussion around AI focuses on whether entry-level software jobs will disappear. The Q2 2026 Dev Barometer points to a different concern. The more significant long-term risk is that developers become highly productive early in their careers without developing the deep understanding required for technical leadership later.

This is an important distinction. AI already enables developers to write code faster, troubleshoot more efficiently, and access knowledge instantly. Those capabilities improve productivity across engineering teams. The challenge is ensuring that productivity is supported by genuine expertise rather than dependency on increasingly capable tools.

As engineers progress through their careers, their responsibilities change. Success becomes less about writing individual pieces of code and more about making architectural decisions, evaluating technical trade-offs, managing system complexity, reviewing the work of others, and aligning engineering decisions with business objectives. These responsibilities depend on judgment developed over years of experience and continuous learning.

The survey suggests that both junior and senior developers already recognize AI’s impact on entry-level work. That is not where the greatest uncertainty lies. The unresolved question is whether developers who learn with AI from the beginning can build the same depth of conceptual understanding as those who first learned without it. The current data does not answer that question, but it establishes it as one of the most important issues for technical education and workforce development.

For business leaders, this has direct strategic implications. Organizations should avoid measuring engineering capability primarily through short-term productivity metrics. Faster code generation is valuable, but long-term competitive advantage depends on people who can make high-quality technical decisions under uncertainty, evaluate AI-generated recommendations critically, and continue learning as technology evolves.

This also changes how companies should think about leadership development. Future engineering leaders need opportunities to strengthen systems thinking, technical communication, decision-making, and independent analysis throughout their careers. AI should accelerate these capabilities by reducing routine work.

Organizations that successfully balance AI adoption with continuous capability development are likely to build stronger technical leadership pipelines over time. As AI takes over more repetitive engineering tasks, the value of human judgment, strategic thinking, and cross-functional decision-making will continue to increase rather than decline.

The competencies the technology industry values most have remained remarkably consistent despite rapid advances in AI. Critical thinking, problem decomposition, systems-level understanding, and sound technical judgment continue to define high-performing engineers. AI changes how these professionals work, but it does not eliminate the need for these enduring capabilities.

For executives, the implication is straightforward. AI strategy should not be treated as a standalone technology initiative. It should be integrated with talent development, leadership planning, and organizational learning. Companies that invest only in AI adoption may improve efficiency in the near term. Companies that invest in both AI and durable human capabilities will be better positioned to adapt as technology continues to evolve.

Recap

AI is changing software development at an extraordinary pace. The organizations that benefit most will not be the ones that simply deploy AI faster. They will be the ones that develop people who know how to use AI with sound technical judgment.

The findings from the Q2 2026 Dev Barometer point to an important reality. The market is not placing the highest value on prompt engineering or the latest AI platform. It continues to reward critical thinking, systems understanding, problem decomposition, and the ability to evaluate complex technical decisions. These competencies have remained valuable through every major technology shift because they enable people to adapt as tools evolve.

For business leaders, this should influence more than hiring decisions. It should shape how engineering organizations develop talent, evaluate performance, and build leadership pipelines. Productivity metrics will become less meaningful as AI raises the baseline for everyone. The stronger differentiator will be the quality of decisions engineers make when problems become ambiguous, systems become more complex, and AI-generated recommendations require careful evaluation.

This also presents an opportunity. Organizations no longer need to choose between investing in AI adoption and investing in people. The greatest returns will come from treating both as part of the same strategy. AI should remove repetitive work, accelerate learning, and increase engineering capacity. Human development should focus on the capabilities AI cannot reliably replace: reasoning, judgment, communication, collaboration, and the ability to make informed decisions under uncertainty.

The companies that build these capabilities consistently will create engineering teams that remain resilient regardless of how quickly AI evolves. Today’s junior developers will become tomorrow’s technical architects, engineering managers, and technology executives. The quality of that future leadership will depend on the learning environments, expectations, and development opportunities organizations create today.

AI is transforming software development. The organizations that lead over the next decade will recognize that the technology itself is only part of the equation. Long-term competitive advantage will come from combining increasingly capable AI with people who know when to trust it, when to question it, and how to turn it into better business outcomes.