What changes when you leave the chat tab

Hagar has spent five modules getting comfortable with prompts inside the ChatGPT and Claude browser tabs. Now her side gig — building a customer-support assistant for Bayt Coffee, a small Cairo specialty roaster — pushes her into the API. She is about to learn that the tab she is used to is doing more work than she realised.

Module 6 walks Hagar through the four shifts:

A chat browser tab is not just a "blank box" to type into. It silently bundles three things you never wrote yourself:

1. A hidden system prompt. ChatGPT and Claude.ai both ship with a long, vendor-written system prompt that defines tone, refusals, formatting defaults, and disclaimers. You never see it, but it shapes every reply.
2. The conversation memory. Every previous message in the tab is replayed on every new turn. The model "remembers" because the UI re-sends the whole transcript.
3. Sensible default settings. Temperature, max tokens, model variant, even the system prompt's "be helpful" framing — all picked for you.

Move that same prompt to the API and none of those three things follow you across. You get a clean room. The model has no idea you are building for a Cairo coffee shop. It has no memory of yesterday. It has no built-in "be friendly" instruction.

That sounds like a downgrade. It is actually the entire point. The API gives you three slots that the chat tab hides:

Slot	What it does
system	The sticky instruction. Every request starts with it. This is where you tell the model who it is.
messages	The conversation, including any history you choose to replay.
temperature	How adventurous the model is. Low = deterministic. High = creative.

The skill you have been building — writing a tight prompt — graduates here from a one-shot trick into a permanent contract between you and the model.

Next: a closer look at the system slot — the single most important field in the API. :::