Every AI word, explained with one story: you just hired a brilliant assistant who started this morning.
Meet Sam. Sam has read almost everything ever written, is scary fast, and is happy to help. Sam has also never heard of your company, can't see your inbox, and will confidently guess when unsure. Every word below is about one of three things: what Sam is, what you give Sam, or what you let Sam do.
Sam. Software that learned from examples instead of being told rules.
Sam's brain. There are many Sams; each one is a different model, and new ones come out every few months.
The kind of brain that's good with words. That's the Sam you chat with.
How long Sam thinks before answering. Same Sam, more or less time at the whiteboard. Turn it up for hard problems; leave it down for quick drafts.
Sam's cousins who are good with pictures, sound, or video instead of words. Multimodal means one Sam who can do several.
Whatever you say to Sam.
Whatever Sam says back. It's made up fresh every time, so ask twice and you might get two slightly different answers.
The word-pieces Sam counts in. Long documents cost more tokens. Only matters when you hit a limit.
Sam confidently making something up. Not lying, just filling a gap. Happens most when Sam doesn't have the facts.
The day Sam stopped reading. Anything after that, Sam has to look up or be told.
This is where most of the magic is. Not in how you ask.
Everything Sam can see right now: your message, the conversation, any files, any saved notes. If it's not in view, Sam doesn't know it.
A sticky note on Sam's desk that's there every day: who you are, how you like things, what to never do.
A folder for one area of your business. Open the folder and Sam already has the background. No re-explaining.
The office Sam works in. Who's on the team, what Sam is allowed to plug into, whether your files stay private. The settings that gate everything else.
Things Sam remembers about you between conversations. You can read and erase the list.
Making Sam answer from your documents instead of from general knowledge. "Look it up in this before you answer."
A how-we-do-it-here sheet Sam reads when a matching job comes up. Your SOP, written for Sam.
Something Sam can use besides talking: search the web, read a file, send an email.
A key to one of your services: Gmail, QuickBooks, Drive. You hand it over once; after that Sam can look inside.
The shape of the keyhole. A standard so any key fits any Sam. You'll see the letters; you won't need to touch it.
How software talks to software. It's what's behind every connector. Comes up when someone builds you something custom.
Read: Sam can look. Write: Sam can change things: send, post, pay, delete. Give read first. Give write one thing at a time.
Sam working a whole task alone: figure out the next step, use a tool, check the result, keep going. A chat answers; an agent does.
Sam doing all that with nobody checking. A dial, not a switch. Turn it up slowly.
You, checking before anything real happens. Where you should be for anything that sends, pays, or posts.
A job you've written down so Sam does it the same way every time.
A workflow on a timer. "Every Friday at 4, check this." Good for things you forget. Bad as a weekly report nobody reads.
A workflow that starts when something happens: an email arrives, a form comes in.
The lines Sam won't cross. Some built in, some you write. "Drafts only" is a guardrail.
A stranger hiding instructions in something Sam reads, hoping Sam obeys them. One more reason Sam shouldn't be able to send email unsupervised.
Everything around Sam: the sticky notes, the folders, the keys, the timers, the rules. Sam is the engine. The harness is the rest of the car. When AI works great for someone else and not for you, it's the harness, not Sam.
Don't get better at talking to Sam. Get better at setting up Sam's desk.