We created assembly code so we did not have to remember all the numbers that make up machine code instruction sets.
We created programming languages so we could forget the specifics of proprietary CPUs and operating systems.
Are we now adopting LLMs and MCPs so we don’t have to learn programming languages and APIs?
If so, before too long, we will forget how to think for ourselves and the AI models (and the Tech Bros behind them) will become our overlords.
Top comments (2)
In the Thoughworks article Should we still design code for humans? by Valentina Servile, the convincing case is made to refute my position.
In the Thoughtworks post Valentina recognises the same pattern (and more) as I did but made the argument against my article is that AI tools are non-deterministic. The mapping of assembly to machine code and high-level language into binary are traceable between the higher and lower abstractions in a deterministic manner, but this cannot be said of code generated by AI/LLMs from natural language.
I do like Valentina's suggestion that in order to define the specification with sufficient precision and absence of ambiguity, one must transcribe more text than one would in a conventional programming language.
What to teach when AI writes the code | Rainer Stropek | TEDxLinz