The Current State of AI-Driven Software Development
- The current state of AI-driven software development is described as a process where someone defines the project's requirements, generates a plan, and then iterates with multiple agent instances until it's done, with a growing distance between the orchestrator and the generated code 10s.
- A new approach to computer programming has generated excitement among developers, with one developer claiming that with Claude Code, they have become a 10x developer, and sometimes it feels like 100x, finding it thrilling and amazing to watch the AI work 1m42s.
- This vision of AI-driven software development points to a massive change for the world of software development, where no one would need to learn to code again, and English would be the new abstraction layer, with millions of well-paid and highly skilled software developers being replaced by a small number of orchestrators who manage hordes of coding agents 2m6s.
- However, not everyone believes in this vision, with Lars Fay arguing that this new style of software development isn't sustainable, and that agentic coding is a trap, as it requires skilled developers to think critically and operate at the architectural level to spot issues in generated code 3m30s.
The Vision of a Future Without Traditional Coding
- Fay's main issue with the current vision is that the use of AI tooling can negatively impact the critical thinking skills and cognitive clarity of developers, which are essential for using AI effectively, and this is not an abstract fear, but a reality that programmers are starting to talk about 5m10s.
- To support this claim, Fay points to posts from Reddit where people are worried about losing their ability to code due to AI, with one post stating that the writer feels like their skills are deteriorating, and that not understanding the syntax is a problem, and that checking the AI's work takes a lot of time 6m30s.
- Established developers are experiencing a decline in their coding skills due to the increasing use of AI, with some feeling like they are losing their skills and becoming worse professionals as a result of relying on AI to do their work 10s.
Junior and Senior Developers Struggling with AI Reliance
- Junior developers are also being affected, as they are struggling to write code from scratch and are instead relying on AI, which means they are not developing the necessary skills to understand and review code, with some being unable to debug code they did not write 1m42s.
- The "junior year wall" effect, which describes the struggle that CS students face when they have relied on AI to complete basic coding classes and then struggle with more challenging classes, is now being seen in the entire coding industry, with junior developers being unable to debug code without the use of AI 2m6s.
- A veteran developer with 30 years of experience shared four points about the current state of affairs in programming, including that AI tools can get the job done faster for highly skilled developers, but also lead to context switching and attention fragmentation, skill atrophy, and the potential for management to use token counts as a misguided productivity measure 4m10s.
The Risks of AI in Developer Skill Atrophy
- The use of AI coding tools is not necessarily the problem, but rather how they are integrated into software development, with the need for developers to think carefully about how to use these tools in a way that does not lead to skill atrophy or burnout 8m40s.
- The solution is not to abandon AI tools entirely, but to find a way to use them in a way that complements human skills and abilities, rather than replacing them, with developers like Fay and the veteran developer still using these tools, but being mindful of their limitations 10m20s.
Responsible Integration of AI in Software Development
- Lars Fay suggests demoting AI's role in coding, instead leveraging Large Language Models (LLMs) and coding agents as secondary processes to avoid sacrificing individual skills for productivity 10s.
- Fay uses LLMs more for specifications and planning, and less for coding, writing 20% to 100% of the code himself depending on the task's importance, and often provides pseudo code when having the LLM write code 42s.
- When using LLMs to write code, Fay never asks them to implement something he has never done before or couldn't do on his own, emphasizing the importance of human skills in coding 2m6s.
A Balanced Future of Human-AI Collaboration in Coding
- The vision for the future of coding is one where AI coding tools are ubiquitous, making software development faster and cheaper, but still requiring humans to learn and maintain their programming abilities to review code and work on complex features 2m6s.
- This vision is not a world where computer code doesn't exist or where the number of software developers drastically decreases, but rather one where humans and AI collaborate to improve software development 2m6s.
- The fact that Fay's warnings are attracting attention in the software industry serves as a reminder that the reality of AI's impact on coding is likely to be more complicated and uneven than predicted by enthusiasts 2m6s.
