LLMs Are Still in the MS-DOS Era.

Interacting with today's large language models often feels like stepping back into a command-line interface. You type an exact command. You get output. If you phrased it wrong, you get the wrong output. No undo. No GUI. No intuitive feedback loop.

This is the MS-DOS era of AI. The tools are powerful — genuinely transformative. But the interfaces are still primitive. We're typing prompts the way early computer users typed commands: carefully, with awareness that the machine is literal and unforgiving.

What comes next — the "Windows era" of AI — will feel radically different. Interfaces that understand intent, not just instruction. Systems that ask clarifying questions instead of guessing wrong. Workflows where the AI handles the formatting, the structure, and the edge cases so you can focus on what you actually want to say.

We're not there yet. But knowing we're in the MS-DOS era is itself useful information. It means the awkwardness isn't permanent. It means the skills you're building right now — prompt engineering, context management, output verification — are the early skills that will compound into something more powerful as the tools catch up.

The command line was never the destination. It was the start.

---

*Originally published on [Substack](https://venkatavemuri.substack.com/p/llms-are-still-in-the-ms-dos-era).*