Why this list exists
Words decay. The ones below were useful for a moment, then a vendor stretched them past meaning, then everyone repeated the stretched version, then the term came to mean "I want this to sound impressive" and nothing else.
We don't say these in client meetings anymore. The replacement is always more boring and more correct.
1. "AI-first" / "AI-native"
What it meant: built around AI from the start.
What it now means: nothing — every product is
"AI-first" in 2026.
We say instead: "the AI is on the critical path"
or "the AI is optional." Different products. Be specific.
2. "Agentic"
What it meant: the model can call tools and run
multi-step workflows.
What it now means: "we want it to feel
autonomous, please don't ask exactly how."
We say instead: "it makes 3–5 tool calls per
task" or "the model owns the orchestration loop." Numbers and
verbs beat the adjective.
3. "Copilot"
What it meant: assistive AI that augments a
human worker.
What it now means: any product with a model in
it.
We say instead: "drafts and the human approves"
or "suggests and the human chooses." That's the actual contract;
"copilot" obscures it.
4. "Hallucinate"
What it meant: the model generates plausible-
sounding incorrect content.
What it now means: a polite excuse for shipping
without an eval.
We say instead: "the model was wrong." Models
don't hallucinate any more than your last hire did. They got
it wrong. Treat it like a bug.
5. "Knowledge base"
What it meant: the source of truth a model
retrieves from.
What it now means: "we dumped some PDFs in a
vector DB and called it knowledge."
We say instead: "the documents the model
reads." Specifying which documents, in what state, with what
curation, is most of the work.
6. "Hyper-personalised"
What it meant: tailored to an individual.
What it now means: templated with a first-name
merge tag and a fact looked up from LinkedIn.
We say instead: describe the actual variability
— "the email references the prospect's last podcast appearance and
adapts the CTA to company size." Then you can argue about whether
it's worth doing.
7. "Augmented intelligence"
What it meant: AI that helps humans rather than
replacing them.
What it now means: AI being sold to a buyer
worried about replacing humans.
We say instead: "the human is in the loop on
every decision" or "the human reviews 1 in 10 decisions." Pick
the actual ratio. The phrase hides the ratio.
8. "Foundational" / "foundation model"
What it meant: a large pretrained model that
others build on.
What it now means: "we're using one of the big
models, please don't ask which."
We say instead: name the model. Claude Opus 4.7,
Sonnet 4.6, Haiku 4.5, Gemini, GPT — the choice is load-bearing
and the buyer often discovers later that it matters.
9. "Self-healing"
What it meant: the system detects errors and
retries.
What it now means: "we have retry logic and we
will not be discussing the failure modes."
We say instead: describe the retry policy and
the circuit breaker. "Self-healing" sounds like the system is
mature; specifics show whether it actually is.
10. "Enterprise-ready"
What it meant: SOC 2, SSO, audit logs, role-
based access.
What it now means: "we want to be in
enterprise deals next quarter."
We say instead: list the certifications you have.
If you don't have them, say "we'll be SOC 2 by Q3, and here's
what we have today." Skip the umbrella term.
11. "Human-in-the-loop"
What it meant: a human reviews or approves the
AI's work before it has effect.
What it now means: "a human can theoretically
intervene at some point if they notice."
We say instead: describe the approval flow. Is
the human in the loop on every output? Sampled? Only on
flagged outputs? The phrase is true at every level and useful
at none.
12. "AI strategy"
What it meant: a coherent plan for using AI in
the business.
What it now means: a deck.
We say instead: "what AI features ship in the
next two quarters and what they cost." That's the strategy. For
smaller companies specifically, see our companion piece on
what 'AI strategy'
means when you're under $10M ARR.
The replacement rule
Every retired phrase has the same fix: replace it with what's actually happening in the system. Numbers, names, ratios, verbs. The phrases died because they were optimised for sounding-good in a meeting. Specifics are optimised for being true after the meeting. The latter is what gets shipped.