Every AI system you deploy becomes a small character in your life or your organisation's workflow. The name you give it shapes how people relate to it — how much they trust it, how much latitude they give it, whether they forget it exists or address it out loud without thinking.
What the name carries
A name announces expectations. Clippy announced helpfulness and ended up announcing intrusion. Alexa announced familiarity and ended up confusing anyone who shared the name. Siri — Old Norse for "beautiful woman who leads you to victory" — landed somewhere between competent secretary and minor household oracle. None of these outcomes were fully predictable from the name alone. None were accidents either.
The worst AI names are aggressive acronyms: JARVIS, ARIA, NOVA, anything where someone worked backwards from the initials. The problem isn't phonetics — it's that acronym names signal the name was chosen by a committee that had no strong feeling about what the thing actually was. The entity ends up inheriting the committee's ambivalence.
What to look for
Etymological depth. Names that come from somewhere — mythological, literary, historical — carry weight without explaining themselves. Mnemo, from Mnemosyne (the personification of memory in Greek myth), requires no annotation: you hear it and understand what it does. Borges, after the Argentine fabulist who wrote labyrinths of text, is immediately legible to anyone who's read him and pleasantly mysterious to anyone who hasn't.
Phonetic distinctness. One syllable is too short for most spoken contexts. Five is too long for anything said aloud twenty times a day. Two or three, with a consonant that registers when called across a room, is the working range. Hard stops (K, T, X) carry better than sibilants (S, SH, Z) in ambient noise.
Surname and product-family compatibility. If the assistant lives inside a company or a suite of tools, say the name alongside five surnames or product names from different language backgrounds. Listen for the combination that suddenly sounds wrong. A name that clashes in three out of five contexts becomes a running awkwardness.
Distance from the current moment. The top-fifty baby names of the past decade make weak AI names. The tool is not a person and doesn't need to pass as one. The names that date fastest are the ones that were trying hardest to be human-but-not-human at a specific historical moment — familiar enough to seem approachable, novel enough to seem modern, and neither for long.
How we approached it
The AI assistant category in pickme's curated list was built around one constraint: every name must have a one-line origin that makes it feel earned. Mnemo from Mnemosyne. Borges the fabulist. Tycho, the astronomer who wore a prosthetic silver nose and kept a pet moose. Each entry is a small argument for why that name, in that register, works for an entity whose job is to hold knowledge and return it on request.
If none of them land right for your project, the smart generator takes a domain, a personality, and a reference figure and returns suggestions from four models in parallel — sometimes what you need isn't a list but a direction.
One last thing: give the name a month's trial before you commit. Say it in a meeting. Say it when the system fails. Say it to someone who'll use it every day. The name you'll keep is the one that doesn't make that last person wince.