I was watching Braveheart again when the line "pay homage" hit me. I knew what the scene was doing, but I realized I was not even sure how to spell the word. I went to look it up expecting a quick spelling check and found a lot more baggage than I expected.

One meaning is familiar and generous. Homage can be a tribute, a nod of respect, or a visible influence. We say a director paid homage to an older film. We say a musician paid homage to a genre that shaped them. It sounds creative and open. The older meaning is much heavier. In the feudal sense, paying homage meant pledging loyalty to a lord, accepting rank, and acknowledging authority. Same word, different posture.

That split meaning feels especially relevant right now because we are building a new relationship with language through large language models. In earlier software eras, we mostly translated our intent into clicks, forms, or code. With LLMs, we express intent directly in words. The interface is prose. The command line is a sentence. The feature request is often a conversation. If language is the interface, then vocabulary is not decoration. It is infrastructure.

This is where many teams get tripped up. We write a prompt that sounds clear to us, but the model reads different intent from the same words. Ask for a "summary" and you might get compression, opinion, or accidental omission depending on context. Ask it to "analyze" and you might get pattern spotting, critique, or fictional confidence. Ask it to "recommend" and it may optimize for plausibility, not truth. The gap is rarely about intelligence alone. It is usually about definitions that were never made explicit.

The practical shift is subtle but real. We are moving from only doing work ourselves to also directing systems that can do portions of the work with us. That changes what skill looks like. Skill now includes specifying constraints, defining done, setting risk tolerance, and checking outputs with discipline. In other words, we are not replacing expertise with prompts. We are expanding expertise to include prompt clarity, review habits, and operational judgment.

I see this most clearly in writing and coding workflows. If I ask an LLM to "clean this up," I get style polish. If I ask it to "preserve all claims, remove redundancy, reduce reading level to grade 9, and keep the same tone," I get a controlled transformation. If I ask for "a test for this file," I might get something that compiles but misses behavior. If I ask for "a regression test that fails on current code and passes after the fix for case X," I usually get something valuable. The difference is language precision, not magic.

There is also a social side to this transition. Teams now share prompt patterns, model settings, and review checklists the same way we once shared shell snippets and SQL recipes. That is healthy, but only if we stay honest about what words mean in our context. "Safe," "done," "accurate," "high confidence," and "production ready" need explicit definitions. Without that, people think they agree while quietly working from different maps.

This is why that old word stuck with me. Homage can be respect, and it can be submission. In the AI era, we should absolutely respect these tools for how much leverage they provide. We can learn faster, prototype faster, and communicate ideas in ways that were tedious before. But respect is not surrender. We still own the standards, the ethics, and the final judgment. Paying homage to a tool should never mean handing it authority we did not intend to give.

Even pronunciation carries this tension. "HOM-ij" feels practical and contemporary. The French-leaning pronunciation sounds ceremonial. Both are valid in everyday speech, but neither should distract from the deeper point. Language is doing real work now. It is shaping outputs, influencing decisions, and setting quality bars. The more powerful the model, the more costly vague language becomes.

So maybe the lesson from that Braveheart moment is not about pronunciation at all. It is about intent. Are we using words to direct a capable assistant with care, or are we casually yielding authority because the output sounds fluent? Are we treating language as a precision tool, or as a soft suggestion? Which LLM do you pay homage to?