Every bot sits on top of a model. If you're going to manage bots, it helps to know what that engine is and what it isn't.
The short version
A model is a very large pattern-matcher for language. It was trained on a huge amount of text. From that, it learned how words, ideas and steps tend to follow each other.
When you give it a message, it works out a likely good response, one piece at a time. Those pieces are called tokens. A token is roughly a word or part of a word.
That's it. There's no database of facts inside it that it looks up. There's a learned sense of how things usually go.
Why that's powerful
That learned sense turns out to cover a lot. A good model can:
- read a messy email and pull out the three things that matter
- write a first draft of a procedure
- reason through a problem step by step
- decide which tool to use next
Its intelligence is similar to a person's in useful ways. It can follow instructions, weigh options and make judgment calls. It is not the same as a person's, and the differences are where management comes in.
Why that's a risk
Because a model predicts what a good answer looks like, it can produce something that looks right and isn't. That's where hallucinations come from. It isn't lying. It's filling a gap with the most likely-looking thing.
A model also has a cutoff. It learned from text up to a certain date. Anything after that, it only knows if you tell it or give it a tool to look it up.
A real example
xAI's docs list the details for each Grok model. For example, Grok 4.7 has a 500,000-token context window and a knowledge cutoff of May 2026 (docs.x.ai). So if you ask it about something that happened in September 2026, it has to search or be told. Otherwise it's guessing.
Model vs. the bot on top of it
| Model | Bot | |
|---|---|---|
| What it is | The trained engine | A persistent worker built on a model |
| Remembers you tomorrow? | No | Yes, through saved memory and files |
| Has a job? | No | Yes, written down |
| Has tools? | Only what you hand it | A set of tools and connectors |
A model is a bit like raw talent. A bot is that talent hired, trained, given a job and a desk. See What isn't a bot for the full ladder.
What this means for you as a manager
- Don't treat output as fact. Treat it as a capable first draft. Check what matters.
- Give it the facts it needs. A model that's handed the right document beats a smarter model that's guessing.
- Write things down. The model won't carry your instructions forward on its own. The bot's files do that.
The rest of this site builds on this one idea: a capable engine that predicts well, wrapped in a system that keeps it honest.