codemanship in the article Why LLMs Will Always Need An Expert In The Loop (published February 14, 2026):

So, there are two very likely futures we can work with (or, more accurately, around). We should not expect LLMs to get significantly more reliable, and we should not expect them to make sound decisions about modular design, dependencies and high-level architecture on non-trivial systems.

The full skinny: a software developer will be required behind the wheel for any “AI” or “agentic” workflow that’s built around LLMs.

A software developer yes, but where in the process are they needed. There is lots to be done in tooling yet. Lovable is a good example, by setting the guardrails on the platform it enables less technical users to achieve great things.