Is Coding 'Solved' by AI? A Developer's Perspective on AI's Impact and Pitfalls

A contentious claim by Boris suggested that Claude Code had rendered engineers obsolete, sparking a widespread debate on whether AI has truly 'solved' coding. This bold assertion initially met with skepticism but has gradually shifted perspectives as developers witness AI's growing impact on their daily workflows. Understanding the nuances of this debate requires acknowledging the diverse interpretations of 'solved' and 'coding' within the AI development community, moving beyond superficial discussions to address the actual capabilities and limitations of current AI agents. The speaker notes, "Coding is solved. Bugs are not yet solved. Fix incoming."

A specific bug found in the Claude Code desktop application illustrates a common flaw in AI-generated code: the inability to fully grasp user interface context. The app displayed truncated update text, a visual defect that would likely have been caught by human verification. This issue highlights a key difference from pre-AI development, where such UI-related errors were less prevalent in code reviews. The speaker points out, "This is such a common mistake that I see all the time with AI generated code that frankly didn't happen a whole lot before AI generated code took over."

The rise of AI in development is leading to a concerning shift in code review culture, where the submission of a pull request no longer guarantees that the code has been thoroughly tested or even manually verified. This new paradigm creates an environment where developers might mistakenly assume that AI-generated code is functional simply because it exists, without critical human oversight. The speaker emphasizes, "But now we live in a world where the code review going up does not imply any effort was put into checking the thing ahead." This erosion of developer quality assurance standards directly impacts the end-user experience, moving the problem from 'coding being solved' to 'functional UI' being an unaddressed challenge.

To have a meaningful discussion about whether AI has 'solved' coding, it is crucial to establish clear definitions for terms such as 'coding' and 'solved.' The speaker raises questions about the definition of a 'bug' when features are visually cut off on a screen, highlighting that such issues might not be considered traditional bugs but are still functional problems. The definition of 'coding' itself varies widely, from simple text editing to comprehensive end-to-end product development, making a unified understanding essential for productive dialogue. As the speaker states, "In order for us to have a conversation about this, we need definitions for the words coding and for solved."

The speaker uses Neil Fun's 'Password Game' as an example to distinguish between simple coding and actual engineering. This game is intentionally designed to be frustrating, mocking the overly complex password requirements prevalent on many websites today, which often lead to a poor user experience. The illustration demonstrates that both intentionally bad design and accidentally poor implementation are fundamentally acts of coding. However, engineering involves a deeper consideration of user experience and overall system health, not just the creation of functional code.

A successful engineering cycle hinges on both sound planning and effective execution, but engineers often face pressure to implement product team plans rather than challenge them. The core question then becomes whether an exceptional coder, whose job is to translate ideas into code, would push back against a flawed plan or merely execute it because of their coding proficiency. The speaker argues that an exceptional coder is distinguished by their ability to critically evaluate and challenge bad plans, rather than simply implementing them. The emphasis is on making good software, not just good code, because even great execution of a bad plan will inevitably lead to bad software. The speaker states, "Is a truly exceptional coder, a person whose job is to turn ideas into code, are they going to push back on a bad plan or would they just implement that bad plan because they're good at coding, not making good software?"

The quality of a plan directly correlates with the success of a software project. Good plans, when executed by skilled engineers, typically result in high success rates, estimated between 8 and 10 out of 10. Conversely, bad plans almost invariably lead to poor outcomes, often as low as 1 to 4 out of 10, regardless of the engineers' capabilities. Furthermore, vague or ill-defined plans introduce significant volatility, making it nearly impossible to predict the final quality or success of the software. As the speaker puts it, "If the plan is good and the engineers are good, we can say the range of success here would be like 8 to 10 obviously out of 10. Where with the bad plan, I would argue that the software coming out will also probably be bad as low as like a one to a four out of 10."

Autonomous agents can generate code and even create pull requests without undergoing actual testing, leading to a critical failure point in the development process. This often means that human users are left to discover and report flaws only after the code has been merged into the main branch. The consequence of merging untested code is a significant increase in the 'friction' required to rectify issues, as fixing problems later in the development cycle is far more resource-intensive than verifying code early on. The speaker warns, "If it hasn't actually run the code to check, it might look fine, it might seem fine, but you don't know if it is fine until someone checks it."

If a developer faces significant difficulty in setting up and testing code locally, autonomous agents will encounter even greater challenges in performing the same verification tasks. Complex and difficult verification processes inherently discourage rigorous testing, making it less likely that code will be thoroughly checked before deployment. Companies that design and build software in ways that complicate verification are the ones most prone to quality issues, as even advanced AI agents cannot compensate for fundamental architectural flaws that impede proper testing. The speaker states, "If it is too hard for you to spin up and test, it is way too hard for your agents to do the same. You need to make it easier to actually verify your code bases."

Software quality is fundamentally dependent on the establishment of a continuous and effective verification loop. This feedback mechanism is essential to confirm that the software functions as intended and meets quality standards. The speaker notes that the Claude Code desktop version struggles with quality compared to its web counterpart primarily because of the inherent difficulty in establishing robust testing and verification processes for a desktop application. He underlines, "that good quality verification loop that is necessary to prove this software works."

The term 'coding' is frequently and often narrowly defined as merely the process of translating plans into lines of code. However, software engineering encompasses a much broader scope, involving the entire development lifecycle from initial planning and architectural design to actual coding and rigorous verification. Therefore, the assertion that software engineering has been 'solved' simply because AI agents can generate code is inaccurate. The ability of AI to write code addresses only a fraction of the responsibilities and complexities inherent in true software engineering. The speaker highlights, "Software engineering isn't just this little piece here where you make code. Software engineering isn't even just this square."

AI agents, much like new human engineers, require access to appropriate resources and a conducive environment to function effectively. Blindly assigning tasks to agents within poorly architected or complex codebases will inevitably hinder their productivity and limit the overall quality of the software produced. Verification remains a critical step; without robust mechanisms for agents to confirm their output, their contributions will be unreliable and require extensive human oversight. The speaker emphasizes, "If a new engineer showed up and they were assigned this change and they couldn't even get the test build running on their machine, much less three of them so they can test your changes too, they're just as screwed as the agents are."

It is imperative that AI agents operate within environments that not only allow but also facilitate the verification of their outputs. Existing software providers must expose specific layers and interfaces that enable agents to validate their work and provide demonstrable proof of their accomplishments. Without these verification mechanisms, agents frequently claim success even when their contributions contain visible failures, necessitating manual correction by human engineers. This highlights the ongoing need for human oversight to ensure the quality and accuracy of AI-generated code. As the speaker states, "We need them to be exposed in ways where our agents can verify their work and prove that they have done such too."

Visually reviewing code is often insufficient for catching many types of bugs, as functional defects frequently remain hidden within seemingly correct code. The traditional method of simply reading code has reached its limits in effectively identifying operational flaws. While architectural issues can still be spotted by human eyes, functional bugs require live execution and testing to be detected. Therefore, a shift towards result-focused code review, where the emphasis is on testing the actual behavior of the software, is necessary. The speaker highlights, "We need to check the results, not necessarily the code. Because let's be real here, nobody would have caught this bug in code review because the code looks fine."

Conflict often arises when individuals hold different definitions of what 'coding' entails, particularly when distinguishing between mere typing and the entire software development life cycle. Boris's provocative claim that 'coding is solved' ignited significant pushback precisely because it clashed with many developers' broader understanding of the term. This disagreement underscores the importance of precise terminology in discussions surrounding AI's role in development. The speaker observes, "That's what the whole coding is solved thing does. Especially if you have different definitions of coding."

Increasingly, top-tier engineers are leveraging AI agents not only for straightforward code changes but also for testing complex theories and exploring ambitious tasks that were previously too time-consuming to undertake. These agents enable parallel execution of ideas, significantly expanding the scope of what individual engineers can achieve. This shift indicates a growing recognition among leading professionals that AI can be a powerful tool to augment their capabilities, allowing them to focus on higher-level problem-solving. The speaker notes, "I am seeing more and more of like the top upper echelon best engineers in the world realizing that they can utilize these agents to do a lot of coding work."

The speaker cites John Ousterhout's framework, which distinguishes between 'tactical' and 'strategic' programming paradigms. Tactical programming refers to the day-to-day, on-the-ground aspects of coding, involving immediate implementation and problem-solving. Strategic programming, in contrast, focuses on the long-term health of the codebase, architectural decisions, and future scalability. While current AI agents are highly effective as 'tactical tornadoes,' excelling at rapid code generation and localized tasks, they largely lack the capacity for strategic thinking required for long-term planning and architectural decisions. The speaker quotes, "He cites John Asterhout's definition of tactical versus strategic programming. Tactical is the on the ground day-to-day aspects of coding, and strategic is the long-term stuff"

Automating pipelines with AI agents requires significant human engineering effort upfront to establish the necessary frameworks and guidelines. Human judgment remains indispensable for managing high-level processes, such as Request for Comments (RFCs) and architectural changes, where complex decision-making and strategic foresight are crucial. This indicates that while agents can handle tactical coding tasks, human oversight and engineering expertise are still essential for designing and maintaining effective, strategic AI pipelines. The speaker states, "In order to make your agents act more strategic and act more like engineers, you have to do the engineering upfront yourself."