'Chatbot' used to mean something narrow: a scripted box on a website answering from a fixed decision tree, the kind that's existed since ELIZA in 1964. That's not really the confusion anymore. Nobody's searching to compare a 2016-era FAQ bot to an AI agent. The actual confusion is that most of what's marketed as an 'AI agent' today is still, functionally, a chatbot: a reasoning one, sometimes a genuinely impressive one, but still something that only does anything once someone messages it. And the vendors selling them will tell you as much, if you read past the headline.
Here's the actual test for spotting a chatbot dressed up as an AI agent, then the receipts.
The actual test
Ignore the model behind it, ignore the chat window, ignore the word 'autonomous' on the pricing page. Three questions settle it: does it do anything if nobody messages it today? Does it have a standing goal, or just a conversation to respond to? And can it be assigned new work by someone other than whoever it's currently talking to?
If the honest answer to all three is no, you're looking at a chatbot. Possibly a very good one mid-conversation. But it exists to answer, not to work.
Chatbot vs AI agent: what the vendors' own pages say
Three of the most prominent 'AI agent' products on the market today are built almost entirely around that same trigger: someone has to message them first.
Salesforce's own explainer on AI agent vs chatbot argues agents reason, decide and execute complex workflows on their own. Its FAQ a little further down the same page gives the game away. Asked whether a chatbot can become an agent, Salesforce says yes, once it picks up 'more advanced AI capabilities like reasoning, planning, and external tool use.' That's Salesforce's own admission that the line is a spectrum of how much reasoning got bolted onto a chat interface, not a different category of product. Its own 'ideal use cases' for an agent, resolving a billing dispute, processing a return, are still requests a customer brought to it, not goals it went looking for on its own.
Zendesk's AI agents page defines them as autonomous systems that resolve customer service requests across messaging, email and voice. Every capability under that definition, reasoning through a multi-intent request, orchestrating actions across systems, closing out a billing issue, happens inside a conversation a customer started. Nothing in Zendesk's own FAQ describes an agent doing something because a target needs hitting, rather than because someone typed a message.
Intercom's Fin is the most honest case, because Intercom splits the difference itself. Fin, described on Intercom's own site as its flagship Customer Agent, resolves conversations across chat, voice and email, still triggered by an inbound message. Sitting next to it is Fin Operator, pitched as a separate layer that finds problems in the support queue and fixes them without being asked. That's the real dividing line, inside one vendor's own product line: Fin answers, Operator works. Only one of the two would pass the three questions above.
Why this matters more than semantics
None of this makes Salesforce, Zendesk or Intercom's products bad. Reasoning inside a conversation is genuinely useful, and calling it an agent isn't really dishonest, it's just this generation's word for 'chatbot with an LLM instead of a decision tree.' The problem shows up when a buyer assumes 'agent' means a thing with a standing goal, working whether or not anyone's watching, and it turns out to mean a very good responder. Buy the wrong one expecting the other and the disappointment isn't the tool's fault. It was never built to do what the label implied, and without something that actually owns the cadence connecting a goal to the work underneath it, StratOps has nothing to plug into.
The fuller argument, and why we draw the line differently
We've made the case for what actually separates a chatbot or an assistant from an agent in more depth in AI agent vs AI assistant. The short version is the same test above: does it own an outcome, or does it just answer questions about one.
The same test, applied to us
Tability draws this line deliberately, including on ourselves. Here's the comparison we show anyone asking what makes a Tability Agent different, and it's the same three questions above turned into a feature list:
| Most "AI agents" (chatbots in disguise) | A Tability Agent |
|---|---|
| Answers questions when you ask | Has a goal it's working toward, so it doesn't wait for your next command |
| Lives in a chat sidebar | Generates its own tasks toward its goal, without being asked |
| Forgotten once the chat closes | Runs scheduled work through connected models and skills |
| Reports to no one | Reports outputs and results in check-ins, comments and dashboards |
| Belongs to whoever opened the chat | Lives in the org chart and can be assigned work by anyone on the team |
Tabby, our AI assistant, sits on the left side of that table on purpose: ask it about your OKRs and it answers from live workspace data, on demand, the same trigger-and-respond model as Fin or Zendesk's AI agents. A Tability Agent sits on the right: give it a goal and it plans the work, recruits whatever specialist agents that plan needs into an Agency, and reports on a schedule, whether or not anyone opened the app today. Once a team is running more than one goal-owning agent, the next question stops being chatbot-or-agent and becomes how to manage the fleet, covered in agent workforce management.
Try it yourself
The fastest way to feel the difference is to hand an actual goal to an actual agent and watch what happens next. Sign up free and assign your first Agent a goal, or book 30 minutes with us if you'd rather watch it work a real key result first.
.jpg)
