Coding with AI: 2 alternative philosophies
Broadly speaking, AI-coding techniques fall into two distinct and alternative philosophies, you don’t want to confuse or mix.
I have listed below common AI-coding techniques, placing them on a spectrum from AI-driven to Human-driven:
Broadly speaking, these AI-coding techniques fall into two distinct and alternative philosophies.
One philosophy is to consider code as scrap, a disposable byproduct.
It is non-technical people who typically embrace this philosophy. They review the outcome and largely ignore the code. The focus is on the prompts used to produce the desired outcome. Agents swarming and Vibe coding are examples of this philosophy.
The other philosophy is to consider code as a permanent asset.
This is because the code’s internal quality influences the external quality of the final product, including how easily it can be operated, used, supported, maintained, and evolved. AI-assisted coding and Inline suggestions are examples of this philosophy.
Today’s LLMs’ limitations still force a choice: hyper-speed or quality
While we try to move toward a future of fully autonomous AI-orchestrated, multi-agent AI code generation, current LLMs are not there yet. Even within a Human-driven approach, like for example AI-assisted coding, today’s LLMs gradually introduce, under the radar, small problems like code inconsistencies and sub-optimal design choices. These small issues accumulate over time, eventually compromising code readability, maintainability, evolvability, and system security, reliability, performance, scalability, etc.
At this stage of the AI revolution, hyper–speed still comes at the cost of quality. Choosing the right AI coding technique and philosophy requires a careful evaluation of the size/maturity/criticality of the product one is developing, and the expectations of its users:
The classic trade-off between hyper–speed (productivity) and quality persists in the code generated with the AI, a fact supported by recent research (see AI-assisted coding: how to ensure real productivity gains?). It is therefore essential to align the AI coding technique used and its philosophy with the trade-offs needed by each product type (from mock-ups to legacy products):
Conclusion: Choose wisely, remain consistent, and don’t overlook quality.
Develop Technical Excellence that delivers.
See how we can help.
You, your team, your Tech.