Talking to Machines: What Linguists Already Know About Context, Meaning, and Skills

"Can you fix it, it's broken again." What does it refer to? Linguists could easily answer this question long before machines could talk back. "Context engineering," AI "skills", and confident-but-wrong AI output aren't new problems - they're classic linguistic phenomena (deixis, presupposition, register, etc.) that the IT/AI industry is rediscovering under new names. Working across a Mediator/Generator paradigm split (i.e. AI as a directed assistant vs. AI producing output with minimal human touch) we'll discuss how the same ambiguous phrase resolves safely when a human is present to ask "which one?" and ships as a silent, costly mistake when no one is. The payoff is practical: a checklist you can use for spotting ambiguity in your own prompts and tickets, and a decades-old aerospace discipline - Controlled Natural Language, or Simplified Technical English - for writing instructions machines (and people) can't misread. This presentation is for developers, PMs, technical writers, and anyone writing docs, tickets, or prompts that AI will act on.