{
  "id": 176085,
  "title": "AI took my job as a coder. Then it made me a better software engineer.",
  "url": "https://urgent.news/2026/08/05/ai-took-my-job-as-a-coder-then-it-made-me-a-better-software-engineer",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-05T13:31:00.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/geewiz/ai-took-my-job-as-a-coder-then-it-made-me-a-better-software-engineer-4519"
  },
  "original_language": "en",
  "account": "I have been using agent-supported coding for a few weeks now, and it has undoubtedly transformed the way I work. It is not worse, just different. This observation is remarkable considering that many programmers are experiencing a drastically different outcome: LLMs generating subpar code, introducing bugs, even ruining data. Additionally, there are reports of skill decay, diminished job satisfaction, and a sense of disconnect from their own codebase. These developers seem to believe that while the LLM fulfills their desires, it simultaneously takes away something precious to them. I do not wish to find myself in this unfortunate situation or face the consequences of skyrocketing LLM prices or the collapse of the AI bubble. My concern is not ending up stranded; it is attributed to my attitude towards coding or, more accurately, towards software engineering. Working with an agent is when good coding standards truly shine. I have experienced this firsthand when we implemented linting, unit, and integration tests into our infra code several years into our operation. We quickly realized how our confidence in the deployed code grew, while the number of emergency fixes during the night decreased. Even with guardrails in place, I do not rely on the agent to navigate independently. My agent sessions resemble pair programming, with the agent as the driver and me as the observer. I witness every step the agent takes. Not only out of lack of trust but also by asking questions about its approach and intervening to help it find the best solution. Each task concludes with a merge request that we meticulously review before integrating the changes into our codebase. My extensive experience has taught me that investing in seemingly mundane engineering practices pays off. I do not see why this should change merely because LLMs are involved. To me, programming and engineering are two distinct concepts; I care about the final outcome, not how it is achieved. Mark Betz, in his article \"My Software North Star,\" concisely states that software should be useful to the end user, correct, maintainable, and efficient. While the methods for achieving this have evolved over the years, their core intent remains intact. Companies now focus on discovery instead of requirements analysis, TDD instead of formal verification, and ADRs instead of extensive design documents. However, these modern approaches are still not significantly lighter for a two-person team. Learning and practicing these engineering methods, let alone employing them regularly, always felt like a luxury our team could not afford. The arrival of AI agents has altered these economic dynamics. Initially, I asked the LLM to implement small new features. I found it amusing how proficient it was at Rails coding compared to me. Soon, I adopted the practice of ending prompts with the question of what additional information was needed. This simple act eliminated much backtracking and repetitive work, leading naturally to the agent generating a comprehensive plan first, which was then implemented in a subsequent phase. Crafting these plans necessitates making decisions, both during their inception and implementation. Recognizing the importance of documenting our technological direction, I began experimenting with Architecture Decision Records (ADRs) a few years ago. ADRs serve as documents that preserve the rationale behind significant decisions affecting the software's structure. Having spent 16 years at the same company, I have come to appreciate the value of organizational memory. Writing these documents, however, often proved burdensome, especially since I was not fluent in the language. Fortunately, a dedicated coding partner maintained a repository for ADRs, allowing me to contribute more frequently. As my proficiency in guiding the coding partner increased, I expanded my role beyond feature-building. I started engaging the agent even before launching new projects. Extensive \"crucible sessions\" involving the agent resulted in clear documentation of a project's motivations, goals, and most importantly, its scope. Maintaining focus has always been a challenge for our team, as there is a constant temptation to address random issues discovered during project work. Rationalizing these distractions as \"good opportunities\" is all too easy. Establishing a clear definition of a feature and its scope has significantly helped our team avoid such detours. In refining this workflow, I discovered that this type of document is called a Product Requirement Document (PRD). The agent assists in creating them as project documents in our Linear project management tool. A common thread emerges in all these improvements: they preserve knowledge and foster transparency. This pattern now replicates across all levels of my work. I write less code but create more value. My identity as a software engineer is not tied to writing code or provisioning servers; it is about delivering value. The agent that performs these tasks for me does not diminish my sense of identity. My job has never been solely about delivering code, and I do not feel disenfranchised by the agent's assistance.",
  "summary": "I’ve been doing agent-supported coding for a few weeks now, and it definitely has changed the way I work. It’s not worse, just different. Which is remarkable, given that many programmers are having a disturbingly different experience: LLMs generating bad code, introducing bugs, even destroying data. I also read accounts from people lamenting skill atrophy, loss of job satisfaction, and a feeling…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}