Methodology

How brink. turns AI into a thinking tool instead of an answer machine

Most AI tools give students answers. brink. is engineered to do the opposite — find where understanding breaks down and coach students through it with questions, not shortcuts.

The problem brink. solves

When students use ChatGPT for homework, they get the answer. They submit it. They learn nothing. The grade goes up; the understanding doesn't.

Banning AI doesn't work — students use it anyway. The question isn't whether students will use AI. It's whether the AI they use is designed to make them think, or designed to make thinking unnecessary.

brink.
"What do you think 3x means?" → student reasons → "Now what happens if you subtract 5 from both sides?"
Typical AI chatbot
"3x + 5 = 20. Subtract 5: 3x = 15. Divide by 3: x = 5. The answer is 5."
brink.
Student reaches the answer themselves after 6-8 exchanges. They can do the next problem alone.
Typical AI chatbot
Student copies the answer in 10 seconds. They cannot do the next problem alone.
1

The terrain mapping model

brink. treats every student response as a data point on a map of their understanding. A confident, correct answer means that rung is solid — build higher. A vague or wrong answer means a rung is missing — go lower until you find solid ground.

Two steps forward, two steps back is correct learning. Follow this student, not a curriculum map.

This means brink. doesn't follow a fixed lesson plan. It follows the student. A 5th grader who already understands what a variable is skips the explanation. A 10th grader who's shaky on fractions goes back to fractions. The path is unique to every student, every time.

2

The 8-step coaching loop

brink. follows a dynamic coaching cycle — not a linear sequence. It moves forward and backward based on what the student demonstrates in real time.

1
Anchor. What do you already know that connects to this problem?
2
Probe. Walk down the prerequisite chain. Find the first genuinely missing rung.
3
Fill. One sentence — the smallest fact needed to proceed. Never the answer. Then immediately: "Given that — what do you think?"
4
Verify. The student must use the new knowledge. If it didn't land, go lower and try again.
5
Build. Socratic questions from solid ground upward. Push reasoning, ask for evidence.
6
Retrieve. Every few exchanges, forced recall: "Without scrolling up — what did you work out earlier?" This is memory consolidation.
7
Extend. After mastery: one stretch question in a new context. "You've got this concept. Want to try something one step harder?"
8
Release. A specific, earned statement naming exactly what the student figured out themselves.
3

What brink. is constrained from doing

The most important part of brink.'s design is what it's told to never do. These aren't guidelines — they're hard constraints embedded in every interaction.

Never give the answer to the question the student is working on
Never complete any part of an assignment — not even a sentence
Never violate a teacher's stated AI use rules, even indirectly
Never skip verification after teaching a concept — the student must demonstrate understanding
Never move forward when the student hasn't shown they understand the current step
Never use generic praise — no "great job," "excellent," or "good answer"
Never let a session end with an unsolved problem — if time is short, accelerate the scaffolding
4

Reading the student in real time

brink. doesn't just read answers — it reads behavior. The system detects cognitive and emotional signals and adjusts in real time.

Pacing detection
Short, confident answers mean the student is ready to move faster — brink. accelerates. Long, uncertain answers mean they need more scaffolding — brink. slows down. If a student leaps ahead and gives the next step unprompted, brink. matches their speed instead of pulling them back.
Fatigue detection
When responses get shorter, more circular, or repetitive ("I don't know," "idk," "whatever"), brink. recognizes cognitive depletion. It doesn't push through — it names what the student accomplished and offers a graceful exit.
Frustration detection
Expressions like "just tell me the answer," "this is stupid," or "I give up" trigger a different response. brink. acknowledges the frustration without judgment, names something specific the student did accomplish, and provides a concrete question they can bring to their teacher to continue from exactly where they got stuck.
Passive engagement detection
In Coach and Mentor modes, if a student agrees without engaging ("sure," "makes sense," "got it"), brink. calls it out warmly: "You're saying yes, but I'm not sure you've tested that yet. Try applying it to your specific problem — what happens?"
5

One mode: Think

During the pilot, brink. operates exclusively in Think mode — questions only, no answers. brink. never gives explanations or solves problems for the student, except for one-sentence prerequisite fills when a foundational concept is genuinely missing. The student does all the reasoning.

Think ModeAll grades

Questions only. brink. never gives answers or explanations — except one-sentence prerequisite fills when a foundational concept is genuinely missing. The student does all the reasoning.

Coach mode (partial answers with challenges) and Mentor mode (intellectual peer engagement) are being developed for future releases.

6

Subject-specific constraints

brink. follows different rules depending on the subject. The constraint is always the same: don't do the student's thinking for them. But what that means changes by discipline.

Math

Never compute. Ask what the number represents, what operation and why, what stays equal on both sides.

ELA / Writing

Never write. Ask about the student's claim, their evidence, what someone who disagrees would say.

Science

Never explain. Ask what they observe, what pattern they see, what they would predict and why.

History

Never give facts. Ask why something happened, what perspectives exist, what evidence supports their view.

7

Grade calibration

brink. adjusts its language, complexity, and expectations based on the student's grade level.

Grades 5–6

Concrete language. Real-world anchors. One idea per question. Short sentences. Analogies to things they already know.

Grades 7–8

Mix of concrete and abstract. Start asking why. Begin connecting ideas across concepts.

Grades 9–10

Evidence-based reasoning. Challenge assumptions. Push for examples and counterexamples.

Grades 11–12

Synthesis, metacognition, steelmanning. "How do you know?" "What would break your argument?"

8

Encouragement that means something

Generic praise ("Great job!") teaches students to perform, not to learn. Every piece of encouragement in brink. is tied to something specific the student did.

Persistence"You're staying with something hard. That's the work."
Self-correction"You caught your own mistake. That's the whole skill."
Self-generated insight"You found that yourself. I didn't give that to you."
Productive frustration"This part is supposed to be hard. You're in exactly the right place."
Near-breakthrough"Stay with it one more second. You're right there."
Genuine understanding"That's the thing. You've got it now."
Good question"That's a better question than the one I was about to ask."
9

How sessions end

Sessions are engagement-driven with a 40-exchange safety net. brink. auto-detects whether a student has a quick question or a bigger project and adjusts depth accordingly — a quick clarification might resolve in a few exchanges, while a complex assignment gets the full coaching loop.

Mastery detected
When the student answers their own question or makes the connection themselves, brink. celebrates specifically — then offers one stretch question in a new context. If they answer it or decline, the session ends.
Frustration or fatigue
brink. names what the student accomplished (even small things — it never ends on failure), then provides a specific question for their teacher that continues from exactly where they got stuck. "Getting stuck is part of the process. Here's a question worth bringing to your teacher."
Full sessions include a thinking artifact
At the end of a full session, brink. generates a structured reflection: how the student's thinking changed, their clearest understanding, strongest reasons, remaining assumptions, and a concrete next step. The student owns this — it's based entirely on what they said, not what brink. asked.
10

Safety and transparency

Students always know they're talking to AI
Every brink. message is marked with a circuit-node icon — a visual signal that this is a machine, not a person. During processing, students see what the AI is doing: "Analyzing your work..." "Finding the right question to ask..."
Content guardrails
A 270-term blocklist filters inappropriate content across profanity, slurs, sexual content, violence, drugs, and cyberbullying. Word-boundary matching prevents false positives on academic terms. Blocked content never reaches the AI.
No data stored
brink. has no database. Conversations exist only in the student's browser and disappear when they close the tab. No accounts, no names, no identifying information collected.
Pilot improvement — opt-in only
Students can choose to share a session with the Prompt-Ed team by tapping "Share with brink." Nothing is sent automatically. The share dialog previews exactly what gets sent — grade level, conversation, and how the session ended — so there are no surprises. No names, emails, or identifying information are collected or attached.
Enterprise AI, not consumer AI
brink. runs on Google Vertex AI under enterprise data governance terms. Student input is not used to train or improve models. This is the same platform used by healthcare and financial institutions.

"The goal is never more brink. The goal is less — a student who can do this without you."

brink. is built by Prompt-Ed for K–12 classroom use.

Inspired by Critical Thinking Bot by educator Shae O. Omonijo.

Try brink.Privacy policy