The future of AI-driven development is eXtreme Go Horse (spavel.medium.com)

A provocative take argues the future of AI-driven development isn’t Agile but “eXtreme Go Horse” (XGH) — an intentionally reckless, speed-first methodology that the author updates as XGH+AI to match how LLM-powered coding tools are actually used. Originally a Brazilian anti-process joke popularized in English in 2018, XGH+AI formalizes habits many teams have converged on: do the first thing that comes to mind, ship immediately, outsource problem-solving and tests to AI, never refactor, blame the model when things break, and treat errors as something to react to only when noticed. The piece lists concrete principles — commit-before-updating, anarchic ownership, relying on AI as scapegoat, and privileging velocity over quality — and argues these mesh cleanly with current GenAI behaviors and incentives. For the AI/ML community this is both explanatory and cautionary: LLMs lower friction to generate code and solutions, which can boost short-term throughput but amplifies bugs, technical debt, brittle architectures, and erosion of critical thinking. Key implications include increased maintenance costs, insecure or untested outputs (AI “pretends” tests), team misalignment when schedules and ownership vanish, and risk that productivity gains are illusory if downstream debugging consumes all savings. The essay closes by warning businesses that mass output of low-quality code won’t sustain profits and urges developers to resist outsourcing judgment to subscriptions and to preserve engineering rigor.