2025 is truly an extraordinary year! The year of Vibe? The year of Agents? For many people, including myself, it was a year of career upheaval. For all software professionals, whether you are a product manager, designer, developer, or tester, it was a year of tremendous change brought about by AI Coding Agent. Our work methods have undergone too many changes, whether proactively or passively. @mitsuhiko's year-end review reflects the feelings of many people. It's not just an analysis of technological trends, but also the author's re-examination of the essence of "software engineering" at the turning point in his life of quitting his job to start his own business. Original article: https://t.co/qnv977Dcbc 🚀 Career Turnaround: From a "Stable Large Company" to "AI-Driven Entrepreneurship" In 2025, Armin left Sentry and started his own new company. • Identity Reinvention: This identity shift transformed him from an architect at a mature company back into a founder who needed to personally handle everything. This sense of urgency, "from 0 to 1," prompted him to fully embrace AI. • Efficiency Revolution: To cope with the massive workload in the early stages of a startup, he abandoned his old programming habits and instead positioned himself as an "engineering supervisor for virtual interns." This shift in work style was key to his ability to handle the early development pressures of the new company. 🤖 Programming Paradigms: From "Writing by Hand" to "Toolkits" Armin shared his most profound experience this year: he has almost stopped typing code manually. • Main toolchain: He switched from Cursor to Claude Code and combined Amp and Pi for development. The interplay between in-house development and outsourcing: Interestingly, while outsourcing services are currently popular, the author prefers "building it yourself" due to the explosive growth of AI programming capabilities. For example, he used AI to write a Python and TypeScript SDK generator himself, instead of relying on third-party commercial solutions. ✨ Vibes: Irrational Decision-Making in the AI Era - Experience Failure: Because AI programming has a history of less than a year, the software engineering experience of the past 50 years seems inadequate in the face of intelligent agents. • Intuition-driven: Due to the lack of hard metrics and historical data, developers often have to rely on "feelings" and "atmosphere" when choosing tools. 🧠 Psychological Impact: The "Quasi-Social Relationship" of Human-Machine Relationships Armin candidly expressed a sense of unease: he found himself developing a kind of "quasi-social relationship" with AI tools. • Emotional projection: Even though he knows that the other party is just a probability model, in long-term collaboration, it is difficult for him not to give the AI personality, and even experience emotional fluctuations. • Vocabulary deficiency: He disliked being called an "intelligent agent" because it blurred the boundaries of responsibility, but he could not find a better word to define this existence that lay between a tool and a companion. 🛠️ Lagging Infrastructure: Four Visions for the Future. In his entrepreneurial practice, Armin discovered that existing engineering tools were completely inadequate to keep up with the pace of AI, and he identified areas that urgently needed restructuring: Version control: Git cannot record the thought process of an AI. We need a new version control system that can record "failed attempts" and "hints". • Code review: The existing PR model is designed for humans, and when humans collaborate locally with AI, this iterative process is "invisible" to the team. • Observability: AI excels at handling complex SQL and eBPF, and future monitoring tools should allow AI to directly take over in-depth system analysis. • Combating “junk code”: The open-source community is rife with uncensored AI-generated code, and we need a new social contract to define contribution norms for the AI era.
