Re-embrace Small Steps: Incremental Verification Beats One-Shot AI Coding
What This Is
A practical article on a Koa2 (Node.js framework) login system was published this week. Instead of having AI generate all the code at once, the author broke development into 5 incremental stages: from a /health endpoint to verify the environment, to registering users and encrypting with bcrypt (a password hashing library), and finally to the end-to-end connection of login and session Cookie (a mechanism for the server to identify user identities). Each step added only one capability, verified with the curl command line, ensuring the previous Checkpoint was correct before proceeding. This "contract first, then code" approach turns black-box development into an attributable, transparent process.
Industry View
We notice the industry is course-correcting the "bulk generation" trend brought by AI coding assistants (like Cursor). Having AI write an entire module at once seems highly efficient, but once context is lost or hallucinations occur, the debugging cost will far exceed the writing cost. The traditional agile concept of "move fast in small steps, verify incrementally" has ironically become a critical weapon for controlling code quality in the AI era. However, opposing voices argue that for early trial-and-error projects, overly rigorous incremental verification can slow down prototype output. Sometimes, the unstructured "get it running first, refactor later" approach yields faster short-term results, and over-engineering is actually a burden.
Impact on Regular People
For enterprise IT: When introducing AI coding tools, "incremental verification" engineering specifications must be paired with them; otherwise, the maintenance cost of generated code will easily devour the time saved during the development phase.For individual careers: The engineering ability to decompose tasks and verify them incrementally is becoming a more core career moat than simply knowing how to write Prompts.For the consumer market: A more rigorous development process means the underlying bugs in software delivery may decrease, but consumers may not perceive an acceleration in the initial product launch speed.