Coding by Feel: When Conversation Becomes a Workflow A new kind of programming is spreading in student clubs, startups, and even inside big companies: “vibe coding”. [I] Instead of typing every line, a person describes the goal in plain language, and an AI tool proposes a draft, suggests a different model, or points to a missing file. Some platforms even let newcomers sign up and ship a tiny app before understanding why a loop works. A prompt may be a polite cue in conversation or a strict command in a terminal, and a draft may be a sketch or a first version. The developer steers the process by naming constraints, testing outcomes, and deciding what to keep, and it can feel closer to design than to traditional coding. In this paradigm, the same word can carry two jobs: “run” may mean executing a program or managing a project, while “branch” can be both a code path and a decision path. The convenience is real, yet the speed can hide what the code is actually doing. [II] Vibe coding shines when rapid learning is the point, not perfection. A small team can hit the ground running by letting an assistant generate a basic interface, then allocating time to test the core actions and validate error messages. To stay safe, experienced builders scrutinize outputs, minimize risky assumptions, and keep an eye on data flow, especially when an app touches payments, health, or school records. A simple rule helps: treat every generated change like a hypothesis that must earn trust through tests. Clear tests raise the bar for changes that look fine but fail in edge cases. Many groups mix in short checklists: write one small test before accepting a change, save prompts beside the code so choices are traceable, and reach out to a teammate for a quick code-review when a new library appears. When deadlines pile up, they often prevent messy “fixes” that later cost more than the first idea. [III] Speed still needs boundaries, because every project sits somewhere between play and responsibility. [IV] A prototype can have a short lifespan, albeit a high impact on learning, so rules can be lighter for low-stakes demos than for tools that affect real users. Even playful projects need a mechanism whereby mistakes are noticed early, such as logs, simple monitoring, and a clear way to revert a change. When a team can infer the origin of a bug from a saved prompt and a single “commit” note, debugging becomes faster and calmer. The aim is to build a workflow that protects its inputs and decisions, so progress stays trustworthy and does not diminish as the codebase grows. [Adapted from https://www.vibeclaw.com] |
