Kodelinesoftware engineering
Development practice

We use AI as a development tool. One senior lead developer owns architecture and production code, reviews generated changes, and checks behavior and maintainability before delivery.

Responsibility

AI is a tool; engineering responsibility stays with us

We use AI to assist development. One senior lead developer owns architecture, technical decisions, and the delivered production code. Generated suggestions enter the same engineering process as other changes. Project coordination can be added where the scope requires it; technical responsibility stays with the lead developer.

  • AI supports implementation; it does not take responsibility for the system
  • The lead developer must understand and be able to explain the delivered code
Vibe coding

A working screen is only one acceptance criterion

In this article, we use “vibe coding” to mean accepting generated changes primarily because the application appears to work, without adequately understanding or reviewing the code. The term is used in different ways. We avoid this practice because visible behavior alone does not establish whether business rules, system boundaries, and error paths remain sound.

  • A successful demo cannot establish maintainability
  • Generated code needs review in the context of the existing system
Maintenance and handover

An example: the same business rule in several places

Consider a hypothetical approval workflow. The screens appear to work, but the same approval rule has been duplicated in several places. A later workflow change updates one copy and leaves the others unchanged, producing inconsistent decisions. This is a possible maintenance problem, not client evidence or a property of all AI-generated code.

  • Understanding the affected logic helps locate every implementation of the rule
  • Architecture review checks where the rule belongs and which boundaries it crosses
  • Appropriate checks cover changed rules and affected paths so inconsistencies can be found
Review and verification

Understand, correct, and check the change in context

The lead developer reviews and understands generated changes, corrects them where necessary, and checks their fit with the existing architecture. Verification uses appropriate tests or other checks for the behavior and risk involved. This includes considering affected workflows and error paths, rather than accepting a change solely because its immediate output looks right.

  • Review responsibilities, data flow, and dependencies across the affected system boundaries
  • Check intended behavior and relevant failure cases
  • Keep the logic and technical decisions understandable for maintenance and handover
AI inputs

Personal data, credentials, and secrets stay out

We exclude personal data, credentials, and secrets from AI inputs. This includes passwords, API keys, tokens, and production credentials. These exclusions define a concrete input boundary; they are not a complete GDPR compliance guarantee.

  • Personal data is not submitted to AI tools
  • Access information and secrets are not submitted to AI tools
Start your project

Your new application

Briefly describe your goal – no finished requirements document needed. We respond within one working day. New applications start from EUR 20,000. Budget still open? Choose “Not yet defined”. Planning and a proposal follow paid discovery.