Lesson 2 of 7 · About 3 minutes
Give the agent a useful brief
Separate what the agent should achieve, what it can use, and what it knows for this task.
01 / Learn the idea
The ingredients
Start with a goal you can check: draft a science-club meeting plan for Tuesday, after lessons, with space for 20 students. Those details let us reject a room that is too small, even if its description sounds perfect. A goal states the result; a constraint states a condition the result must satisfy.
A tool is a function the surrounding application makes available, such as reading a timetable or checking room capacity. In function calling, the model requests a function and supplies inputs. The application runs it and returns the result. Naming a tool in a prompt does not create that connection.[1]
Context is the information available for the current decision: instructions, conversation, and relevant results. Applications can keep task state or retrieve stored memory from earlier sessions. Neither means the model remembers everything. Updating this information is different from training, which changes the model's learned parameters.[2][3]
Original school example
A club meeting
Goal: propose a Tuesday meeting. Context: 20 students finish at 3:30. Tool: read available rooms. Constraint: capacity at least 20. Boundary: draft the proposal; do not reserve a room yet.
The application needs a working calendar connection and suitable access. An instruction alone supplies neither.
02 / Explore the flow · Guided simulation
Give the agent what the task needs
A science club needs a room. Change the goal, available tool, and context; watch what becomes possible.
Original fictional scenario. Decisions are authored; calculations and checks run in your browser. No live AI or external action.
↳ A failed check can send the flow back to another action—or to a human.
Step 1 of 3 · Visible system event
Goal
Find a good room. ‘Good’ does not specify capacity.
Inspect the visible trace (1 events)
- Goal
Find a good room. ‘Good’ does not specify capacity.
This log shows authored actions, tool results, and checks. It is not a model’s private reasoning.
03 / Check your understanding
Your agent must choose a room for 20 students. Which missing input would most directly prevent it from checking suitability?
Where these ideas come from
Primary sources, explained in our own words. Research findings and engineering guidance are different kinds of evidence; neither is a guarantee that an agent will succeed.
Supports: Model requests a function; Application executes it and returns results.
Limit: Describes one implementation; this course teaches the concept without requiring Gemini or a live API.
Supports: Current conversation state; Searchable memory across conversations.
Limit: Storage and recall depend on how the application is built; memory is not automatic or unlimited.
Supports: Prompt changes versus parameter updates.
Limit: Used only for the distinction between prompting and training, not its broader performance statements.
Stemtiq original lesson · Sources checked September 12, 2026 · Editorial review pending. School examples and guided simulations are original simplifications, not replications of the cited systems. No research affiliation or endorsement is implied.