Gartner estimates that, among thousands of vendors calling themselves agentic, around 130 actually are. The difference is having initiative and the power to act, not the quality of the response.
What’s the practical test?
Ask what the system does without anyone asking, and what it undoes when it's wrong. An assistant answers when asked; an agent receives a goal, decides the steps and acts — and every action has its reversal defined beforehand.
Why does the inflated count matter to you?
Because it changes the comparison yardstick. Comparing a renamed assistant with an agentic architecture at the same price leads to the wrong decision, and the cost shows up in production.
And at Zthex, how many actually act?
By the strict definition, three of the five act: AILA runs the conversation, AINA executes, AIVA fetches. AIKA classifies and AIRA presents. We’d rather be precise than inflate the count.
Renaming an assistant as an agent is free. Declaring which of your agents actually act costs the difference between looking bigger and being verifiable.