AI Coding Assistants: Faster Code, or Better Software?

Published: 19 July 2026

Reflections on Microsoft's .NET Skills for AI Agents, software engineering, relational design, business rules, and the long term quality of software systems.

While browsing X, I came across the beginning of a longer post by Anton Martyniuk about Microsoft's .NET Skills for AI Agents. It sparked my curiosity, so I searched for Microsoft's official blog to better understand the technology behind it.

Beginning of Anton Martyniuk's X post introducing Microsoft's .NET Skills for AI Agents.

Microsoft's official blog, Extend Your Coding Agent with .NET Skills, explains the concept in more detail and introduces the open source dotnet/skills repository.

Microsoft Blog:
Extend Your Coding Agent with .NET Skills

After reading the official documentation, I concluded that this represents another step in the evolution of developer tools over the past 30 years. Simply put, it is another tool in the programmer's toolbox.

There is no doubt these skills can make developers more productive. However, they also prompted me to reflect on a broader question:

Will AI coding assistants simply help us write code faster, or will they improve the long term quality of software engineering?

Recently, someone shared an observation with me in a direct message that resonated with me:

"They teach just as much as they need."

Reflecting on decades of programming and database development, I have learned that the long term cost of software is rarely writing the initial code. The real cost comes later because systems must be maintained even though their data models, business rules, ETL processes and architecture were never deeply understood.

In my experience, the real gaps often do not appear until an application has been running for a year or two. As the data grows, business requirements evolve and complexity increases, the original design is placed under real pressure. That is when manual processes, workarounds and unnecessary complexity begin to emerge.

This is where I believe AI coding assistants still have limitations. They can understand programming languages and frameworks exceptionally well, but they cannot automatically understand the deeper knowledge embedded within a project, including its business rules, relational data model, architectural decisions and the reasoning behind them. Those aspects still depend on human understanding.

That distinction, in my view, is the difference between writing software and engineering software.

Further Reading