AI has rapidly evolved from a promising but often clumsy coding assistant to a formidable partner capable of generating production-ready code for common patterns. This dramatic shift, particularly evident in late 2025, challenges traditional notions of software development and, contrary to some initial interpretations, actually elevates the importance of robust engineering discipline.
What Happened
For much of 2025, the prevailing sentiment within the software engineering community was that AI-generated code was largely subpar – a "slop" that might forever remain in the realm of prototypes or debugging assistance. This perspective, while reasonable at the time, was decisively overturned by November 2025. The release of models like Opus 4.5 marked a pivotal moment, demonstrating the AI's ability to produce code of comparable quality to that of a median software engineer, often at significantly faster speeds and lower costs.
This breakthrough wasn't an overnight sensation, but rather the culmination of several technological advancements throughout 2025. Precursors to agentic harnesses – systems that wrap Large Language Models (LLMs) in a loop with various tools – began appearing in late 2024, maturing into a robust reality by mid-2025. The integration of enhanced tool use, sophisticated function calling capabilities, and Multiple Concurrent Processes (MCPs) collectively paved the way for general-purpose usability by the end of the year. This underlying technical evolution provided the foundation upon which Opus 4.5 could achieve its impressive coding prowess, fundamentally changing the landscape of code generation.
Image showing a flowchart with "AI-Generated Code Quality" evolving from "Slop" to "Median Software Engineer": image omitted due to site embedding policy; open the original article (Substack) (opens in a new tab) to view it. Photo/source: Substack (opens in a new tab).
Why It Matters
The immediate consequence of AI's leap in code quality is not a license to abandon established software engineering best practices. Rather, it underscores the need for more stringent discipline. The author, coming from a reliability background, highlights that adapting to new realities is crucial. The ability to quickly generate vast amounts of code that is "approximately as good as that of the median software engineer" introduces a new set of challenges and responsibilities for developers and IT organizations.
For individual developers, this means that while AI can accelerate the creation of boilerplate or common patterns, the critical human role shifts towards higher-level concerns: architectural design, complex problem-solving, deep understanding of system interactions, and rigorous code review. The temptation to "push shittiest code straight into production" is amplified if generated code is simply assumed to be correct without proper vetting. Instead, the velocity afforded by AI requires an even more robust and meticulous approach to testing, integration, and validation.
Enterprises must recognize that the increased speed and cost-efficiency of AI-generated code can lead to a significant increase in code volume within their systems. Managing this influx requires fortified Software Development Life Cycle (SDLC) processes. Quality assurance, observability, and robust deployment pipelines become non-negotiable. The "unwholesome zest for lapping up disgusting technical messes" that reliability engineers possess might be tested more frequently, making proactive discipline paramount to prevent these messes from accumulating. Ultimately, the goal is to leverage AI's speed without sacrificing the stability and security of production systems.
Image showing "What Does This Mean for Engineering Discipline?" evolving from "Ditch it!" to "More of it!": image omitted due to site embedding policy; open the original article (Substack) (opens in a new tab) to view it. Photo/source: Substack (opens in a new tab).
What To Watch
The evolution of AI in code generation is far from over. Developers and IT leaders should closely monitor advancements in AI models beyond current iterations like Opus 4.5, particularly their ability to handle more nuanced or domain-specific challenges. The ongoing development of agentic harnesses and tool integration will continue to enhance AI's capabilities, pushing the boundaries of what automated code generation can achieve.
Critically, the industry needs to establish new best practices and tooling specifically tailored for integrating AI-generated code into existing workflows. This includes refining code review processes to account for AI authorship, developing advanced automated testing frameworks, and ensuring comprehensive observability for systems built with AI assistance. The debate around the ethical implications of AI-generated content, including code, also remains a significant area to watch, alongside the technical merits. The next phase will be defined not just by what AI can do, but how responsibly and effectively humans manage its output.