If you spend even fifteen minutes around modern tech people, there is a good chance you will hear a sentence that would have sounded completely unhinged five years ago:
“Let’s add more context, ground the answer, harden the workflow, and maybe make it a little more agentic.”
That is not a sentence. That is a smoothie made of product meetings, cybersecurity, and robot philosophy.
And yet here we are.
The funny part is that most of these words are not actually new. AI did not invent context, prompt, grounding, or hardening out of thin air like a magician with a MacBook. What changed is that AI made these words more useful, more visible, and much more social. They moved out of specialist circles and into normal conversation.
So if you have been wondering whether these words come from AI-generated code or from humans changing how we talk, the most honest answer is: both, but mostly humans first.
AI Did Not Invent Most of the Vocabulary
There are three big buckets.
The first bucket is old words with new AI jobs.
Take prompt. It used to sound theatrical or vaguely computer-y. Now it means the instruction you give a model in natural language. Cambridge even defines prompt engineering as the process of designing prompts that produce the best results. Same word. New career path.
Then there is hallucination. For a long time it lived mostly in medicine and psychology. Now dictionaries include a computing sense too: false information produced by AI. In other words, the word did not arrive with AI, but AI absolutely dragged it into daily office life.
The second bucket is technical words that escaped into normal speech.
This is where context, grounding, and hardening live.
Context is the best example. It used to mean background, circumstances, or “please read the room.” In AI, it became something more operational: the actual information a model gets to work with. By late 2025, Anthropic was writing about context engineering as the next step after prompt engineering. That is a strong sign that the word stopped being casual and became a design discipline.
Grounding sounds like something your yoga teacher would recommend after a stressful week. In AI, Google defines it much more practically: connecting model output to verifiable sources of information. Suddenly “be more grounded” no longer means “touch grass.” It means “show me the receipts.”
Hardening comes from security, not from AI. Microsoft defines it as reducing attack surface by removing extra resources or tightening configurations. But AI coding tools have pushed many more people into building and shipping software quickly, so security words now show up earlier in the conversation. People do not just ask “does it work?” anymore. They ask “did we harden it?”
The third bucket is genuinely new or newly turbocharged terms.
The funniest one might be vibe coding. Collins named it its 2025 Word of the Year and described it as software development that turns natural language into code using AI. That phrase feels very 2025 because it captures a new behavior, not just a new tool. You are not carefully programming every line. You are steering with intent, taste, and a suspicious amount of optimism.
Then there is agentic, a word that existed earlier in psychology and social science, but Merriam-Webster notes that it gained prominence in 2024 in reference to AI. That matters because it shows how AI does not just create brand-new words. It revives sleeping ones and gives them a startup budget.
And of course we also got the backlash vocabulary: deepfake and AI slop. Merriam-Webster defines deepfake as manipulated media designed to misrepresent someone, and chose slop as its 2025 Word of the Year for low-quality AI-generated content. That is an important clue: when a technology goes mainstream, people do not just invent aspirational words for it. They invent insult words too.
What Is Actually Changing in the Way We Talk?
The interesting part is not only that we have new terms. It is that AI seems to be changing the texture of communication itself.
A 2023 study in Scientific Reports found that AI-generated smart replies changed human communication patterns, increasing speed and positive emotional language. That is already bigger than vocabulary. It suggests that AI does not just help us send messages faster; it can subtly reshape tone and interaction.
Then the evidence gets even weirder.
A 2025 paper on human spoken communication analyzed hundreds of thousands of academic talks and podcast episodes and found a measurable increase in words that ChatGPT tends to generate, including words like delve, comprehend, swift, and meticulous, after ChatGPT’s release. Another 2025 paper found rapid shifts in scientific English, with words such as delve, intricate, and crucial spiking in frequency since around 2022.
That means AI is not only giving us new jargon. It may also be nudging our style toward a certain polished, slightly overcaffeinated, very bullet-point-adjacent way of sounding.
This is probably why so many people now describe perfectly normal emails as sounding “a little ChatGPT.” We are getting better at recognizing the rhythm: clear transitions, neat framing, gentle confidence, and a suspicious fondness for phrases like “it is important to note.”
So Where Do These Words Really Come From?
Mostly from humans.
Humans coined some of them. Humans reused many of them. Humans decided they were useful shortcuts for new habits of work. Then AI accelerated the spread because millions of people started interacting with the same kinds of systems, reading the same documentation, and copying the same patterns into writing, meetings, product docs, and code reviews.
So the deeper story is not “AI invented a new language.”
It is closer to this: AI created a giant social pressure cooker for language.
Old words got promoted. Technical words became social. Some awkward phrases became status signals. Some became jokes. Some became accusations. And a few became so common that they now feel inevitable.
That may be the most human part of the whole thing.
Every major technology leaves fingerprints on everyday speech. The printing press changed writing. Television changed performance. The internet changed irony. AI seems to be changing how we describe thinking itself: not just what we know, but how we retrieve, refine, ground, compress, and harden it.
Which is exciting.
Also a little ridiculous.
Because sometimes “please add more context” is a meaningful technical instruction.
And sometimes it is just the 2026 version of “could you explain that better?”
Sources
- Anthropic: Effective context engineering for AI agents - Evidence that context engineering became a named AI discipline in 2025, positioned as a progression beyond prompt engineering.
- Google Cloud: Grounding overview - Defines grounding in generative AI as connecting model output to verifiable sources of information.
- Cambridge Dictionary: prompt engineering - Shows that prompt engineering has entered mainstream dictionary treatment as an AI-specific term.
- Cambridge Dictionary: hallucination - Includes a dedicated computing sense for AI-generated false information.
- Merriam-Webster: agentic - Notes that the word predates AI but gained prominence in 2024 through agentic AI usage.
- Microsoft Learn: Architecture strategies for hardening resources - Useful evidence that hardening comes from security engineering, not from AI itself.
- Collins: Word of the Year 2025 - vibe coding - Evidence that vibe coding became one of the emblematic AI-era expressions.
- Merriam-Webster: Word of the Year 2025 - slop - Strong evidence that AI slop became culturally visible enough to shape mainstream vocabulary.
- Merriam-Webster: Word of the Year 2023 - authentic - Shows how AI also increased pressure around words like authentic, real, and fake.
- Merriam-Webster: deepfake - Definition and origin context for one of the clearest AI-era media terms.
- Scientific Reports: Artificial intelligence in communication impacts language and social relationships - Experimental evidence that AI-assisted replies can change language use and social perception.
- arXiv: Empirical evidence of Large Language Model’s influence on human spoken communication - Large-scale evidence of post-ChatGPT shifts in spoken English.
- arXiv / FLAIRS 2025: Exploring the Structure of AI-Induced Language Change in Scientific English - Evidence that AI-associated word patterns are also showing up in scientific writing.