Part 1 of 3 in The Language of AI Progress
Years ago, a dear friend told me she didn’t really believe in synonyms. Obviously, she understood the concept. Her point was that two words might carry almost identical definitions and still leave completely different flavors in your mouth. Bitter and sour, for example. Close enough to live in the same neighborhood, but not close enough that you’d confuse one for the other. Somewhere among all those technically interchangeable options, she thought, there was usually one that tasted exactly right.
I have thought about that observation far more often than she could possibly have intended. It has followed me into how I write, how I create content, and how seriously I tend to take language in general. Give me fifteen words that supposedly mean the same thing and I will happily spend an unreasonable amount of time deciding which fourteen should be sent home.
The words we choose can shape how clearly we understand whatever sits behind them. Good language helps us separate things we had previously lumped together, notice differences we might otherwise miss, and eventually talk about those differences with other people. Sometimes the stakes are trivial. Sometimes the vocabulary we settle on influences what we see, what we measure, and what we decide to do next. Better language can make a complicated thing easier to understand.
AI gives us a messy case study in how language shapes understanding. The language is still trying to catch up with the thing itself, and the market hasn’t yet settled on how to talk about it. For now, AI adoption has become the default umbrella term for describing where individuals and organizations are in the journey. Fair enough. But the phrase doesn’t arrive with a shared definition, and to my ear, it sounds a little off. In practice, it can describe almost anything from employees simply having access to a tool, to using one here and there, to materially changing how meaningful work gets produced. Two people can say adoption is going well, or terribly, and be talking about different conditions without realizing it.
Broad terms and unsettled definitions leave a strange amount of interpretive work to whoever is listening. In this case, me. When someone says their organization has “high AI adoption,” I still need to know what they mean before I know what I’ve learned. Maybe they mean enthusiasm. Maybe widespread use. Maybe experimentation. Maybe something has genuinely changed. The adjective can get more precise while the noun underneath it stays stubbornly broad. “Poor AI adoption” doesn’t clarify much either. I know whoever said it is dissatisfied, but I still don’t know what condition they’re dissatisfied with.
When the shorthand hasn’t arrived yet
The vocabulary lag is understandable. New phenomena routinely outrun the language we use to describe them, sometimes leaving a lexical gap. For a while, we borrow old words, stretch their meanings, pile on qualifiers, or explain the thing in a whole sentence because the shorthand hasn’t arrived yet. AI feels squarely in that period. The target itself keeps moving as capabilities change and people find new ways to use them, so even useful language may need to evolve along with the work.
Eventually, though, our messy descriptions collapse into something beautifully compact. Doomscrolling. Rage bait. More recently, AI slop. Each arrived after the behavior and gave us a much easier way to recognize it, point to it, and talk about it with someone else. A phenomenon that once required a sentence suddenly fits inside two words, and once the shorthand sticks, it becomes hard to remember how awkwardly we talked around the thing before it had a name.
I love discovering those words. Few things are more satisfying to me than realizing some wild, chaotic thing I’ve been struggling to explain already has a name. I don’t know how to describe the feeling well, but I can almost feel the ooh, I like that or oh, I needed this yesterday. Suddenly I can stop rebuilding the observation from scratch every time I want to talk about it. I can recognize it faster, point to it more cleanly, and carry the idea into the next conversation without dragging the whole backstory along with me.
Once something has a name, we can compare examples, notice patterns, ask whether the same thing is happening somewhere else, and disagree about it with much more precision. Good shorthand gives us something sturdy enough to build the next thought on.
The words we’re reaching for
AI still feels like it is waiting on more of that shorthand. Adoption has become the catch-all, but people are already reaching for other words as they try to describe different kinds of change: augmentation, integration, incorporation, assimilation, even transformation. They overlap enough that you can often swap one for another without losing the meaning, yet each seems to pull the conversation in a slightly different direction. And, predictably, some of them taste much better to me than others.
Augmentation is probably the word I react to most strongly, and I’ll admit up front that this is personal. I hear it and immediately think of augmented reality, which gives the word an unpleasant flavor I can’t seem to shake. “Stephanie, are you augmenting your work with AI?” is just bad. “William, are you incorporating AI into how you work?” lands much more cleanly for me. I know those reactions are subjective, but that is partly the point. Words carry baggage far beyond their dictionary definitions, and two terms that look interchangeable on paper can steer the conversation somewhere different once we actually start using them.
Integration sits slightly better with me, though it comes with its own problems. I tend to hear systems and software stacks before I hear people changing how they work. Ask me about “AI integration” and part of my brain is already looking for relevant API documentation. Incorporation feels more natural because it suggests something becoming part of an existing process. Assimilation gets tossed around. It carries more weight, sure, and I like that it implies the surrounding system has had to accommodate whatever entered it. Even transformation shows up in these conversations, though that one feels so large it practically arrives with its own consulting deck. Each word emphasizes a different part of what is happening, and the one we choose signals which part of the change we think is worth paying attention to.
I have no interest in refereeing a vocabulary contest. If augmentation works for you, use it. If integration is the language your team already understands, great. I care much more about whether the word helps you say something specific enough to be useful. Can two people hear it and picture roughly the same condition? Can you tell whether that condition has changed? Can you decide what to do next? A term that survives those questions earns its place, regardless of whether it happens to be the one that tastes best to me.
The words have to do some work
While I’m agnostic about which word wins, I’m much less patient about how long we’re taking to choose language precise enough to use. The market will settle this on its own, sure, but companies are making decisions now. They are spending money now, setting expectations now, building training programs now, and asking whether progress is happening. Now. I want language that can survive contact with those questions. Something we can define clearly enough to observe, measure, and act on.
Measurement is where broad language starts to cost us. If a leadership team says it wants AI adoption to improve by the end of the quarter, what exactly would count as improvement? The answer affects what goes on the KPI dashboard, at which point someone like me has to make a chart out of whatever we decided “adoption” means this month. It also affects what happens next. A company might buy more licenses, mandate training, celebrate use cases, or redesign workflows, depending on what it believes the problem is. Broad language can make all of those look like “adoption strategies” even though they are responding to different conditions. Before we measure progress or prescribe a response, we need enough precision to know what we are actually trying to change.
Shared language around AI will no doubt get better with time. New terms will stick, old ones will narrow, and some of the distinctions that feel awkward to talk about today will eventually become easier to name. I’d rather have something useful in the meantime, a working vocabulary for the decisions already in front of us, while the broader language continues to evolve, regardless of where the market lands.
I do have a way of separating some of these states that has become useful to me. I’ve broken the broad idea of “AI adoption” into four milestones that make it easier to talk about where someone is today, where they’re trying to go, and whether meaningful progress is actually happening. I’m under no illusion that I’ve found the vocabulary everyone will use, and I don’t particularly care if I have. It gives me a cleaner way to talk about progress now, which is enough reason to put it on the table. I’ll share it next.