The development process consists of several stages: business analysis, system analysis, preparation, coding, code review, testing, and release. Automating the latter stages, such as testing and release, creates a bottleneck in the preparation phase and risks delivering unverified results.
Automating coding and code review shifts the bottleneck to the QA stage. A single QA engineer checking eight features per day limits the overall delivery speed. Furthermore, treating the codebase as a black box complicates incident resolution, which can lead to significant financial losses.
Replacing experienced developers with less skilled staff using AI is ineffective, as AI reduces the cost of specific tasks but does not replace core engineering competence. The author suggests that the most efficient approach is to integrate AI assistants into every role to automate routine tasks.
Implementing AI at each stage of the development cycle could potentially save 21% of total time and reduce development costs by approximately 20%. This approach allows for incremental improvements while maintaining the human oversight necessary for long-term project stability.
The current development model, designed for human-written code, limits the potential of AI integration. Significant acceleration may only occur when programming shifts away from traditional code writing as the primary method of software definition.
