コード知識ゼロから8,700コミット積んだ僕が、269回「過去に戻した」話。AI時代の境界設計とツール使い分けA developer with no prior coding background shares hard-won lessons from 8,700…

A recent post on the Japanese developer platform Zenn documents an unusual milestone: a person who began with no coding knowledge accumulated roughly 8,700 commits by leaning heavily on AI coding assistants such as Cursor. The account matters because it offers a candid, first-hand look at what it actually takes to ship software when large language models handle much of the writing, and because it reframes the discussion away from raw output toward the discipline needed to keep that output manageable.

The headline figure that gives the piece its texture is not the commit count but the 269 git reverts that accompanied it. In Git, a revert creates a new commit that undoes the changes introduced by an earlier one, allowing a developer to roll back a mistake without erasing project history. A revert rate of that magnitude suggests a workflow built around rapid experimentation and frequent correction rather than careful, upfront planning. For someone learning as they go, this appears to function as a safety net: if an AI-generated change breaks something or drifts from the intended design, the author can back it out and try again, treating version control as an undo button for the entire codebase.

The central lesson the author draws is about boundary design, meaning the deliberate decision of where an AI agent is allowed to operate and where it is not. Left unconstrained, AI coding tools tend to touch more files, refactor more broadly, and introduce more side effects than a user intends, which is likely a significant contributor to the high revert count. The remedy the post advocates is scoping work into smaller, well-defined units, committing often, and drawing clear lines around which parts of a project a model may modify in a single pass. This mirrors long-standing software engineering advice about small, atomic commits, but here it is presented as essential precisely because a non-engineer cannot easily audit every line the assistant produces.

Closely tied to boundary design is the theme of tool selection, or choosing the right assistant for a given task. Cursor, the tool that anchors the article and the source blog's focus, is an AI-first code editor built as a fork of Visual Studio Code. It integrates chat, inline editing, and an autonomous agent mode that can plan and execute multi-step changes across a repository. The author's experience implies that different modes and different tools carry different risk profiles: a tightly scoped inline edit is easier to review and revert than a sweeping agentic change, so matching the tool's autonomy to the stakes of the task becomes part of the workflow.

This account sits within a broader industry shift often described as AI-assisted or "vibe" coding, a term popularized in 2025 to describe building software by conversing with a model rather than writing code directly. The surrounding ecosystem has grown quickly, with GitHub Copilot, Anthropic's Claude Code, Windsurf, Replit's agent features, and others competing for the same users. Cursor's parent company, Anysphere, has drawn considerable attention and investment as one of the fastest-growing entrants. Against that backdrop, testimonials from people who are not professional programmers have become an important, if anecdotal, signal of how far these tools can carry a motivated beginner.

The piece also implicitly underscores prerequisite concepts that remain valuable even when a model writes most of the code. Comfort with Git, an understanding of what a commit and a revert do, and the habit of validating changes before trusting them all appear to be what kept the author's project recoverable through hundreds of missteps. In that sense the story is less a claim that coding knowledge is now optional and more an argument that a different, complementary skill set, centered on judgment, scoping, and version control, becomes decisive.

Readers should weigh the account as one individual's experience rather than a controlled study; commit and revert counts are easy to inflate through workflow habits and say little about code quality or the complexity of what was built. Still, as a practitioner's field report, it offers a useful, grounded perspective on how AI coding tools change the shape of the work rather than simply removing it.