AN AALTOES INITIATIVE
WORKSHOP · SESSION 05

Workshop 5:
AI for Studies.

How to study with AI, and how to build your own study tools · September 14 · Startup Sauna

LevelIntermediate
Duration~2 hours
>Prompt
00 / Before we start
// housekeeping

Open your laptop.

  1. 01
    Wi-Fi network · password
  2. 02
    Log in to an AI tool. Claude, ChatGPT or Gemini; free plans are fine for almost everything tonight.
  3. 03
    Find material from a course you're taking right now. Lecture slides, an old exam, an assignment brief. Real material is the whole point.
  4. 04
    Safe space. Contact people: names
>Prompt
01 / Agenda

Tonight.

  1. 01
    How to study with AI three habits, eight tips
  2. 02
    How agents work
  3. 03
    Demos: two student builds + one live build
  4. 04
    Build-along: pick a tip, build it for your course
  5. 05
    Show and tell
>Prompt
02 / Why us
// Helsingin Sanomat · HS Visio · 23.8.2026

Superkäyttäjät antavat
vinkit opiskelijoille

"Super users give students tips." HS asked a few of us how students should actually use AI. Tonight is the long version, and then you build it.

Ada-Sofia Kari, HS Visio, 23.8.2026
>Prompt
03 / The shift
// three years
2023

A chatbot that was sometimes right.

It often got even basic university maths wrong.

2026

It could almost do your master's.

Agents read your course material, write code, grade exercises and plan your week.

So the question isn't whether to use AI. It's how.
>Prompt
04 / The trap
// what not to do

You can outsource the answer.
Not the learning.

  • Paste the assignment in and type "do this for me": you learn nothing
  • Only study what the exam asks: you're fooling yourself
  • Thinking is slow. Skip it, and you slow down your own development
Paraphrased from Heikkinen and Kiirikki, HS Visio, 23.8.2026
>Prompt
05 / What works
// the learning science

Re-reading feels like studying.
Testing yourself is studying.

01 · retrieval practice

Recall, don't re-read.

Pulling an answer out of memory makes it stick far better than reading it again.

02 · spacing

Come back later.

Reviewing across days beats cramming the same hours into one night.

Good AI habits push you into both. Bad ones let you skip them.

Roediger & Karpicke, Psychological Science, 2006 · Cepeda et al., Psychological Bulletin, 2006
>Prompt
Section 01 / Habits

Three habits.
Eight tips.

Give it context · make it teach, not tell · keep your judgment

>Prompt
06 / Habit 1
// habit 1

Give it context.

Stop starting from an empty chat.

tip 01

One project per course

Brief, slides, your notes, grading criteria, all in one place. Claude / ChatGPT Projects, NotebookLM, or a folder.

tip 03

Old exams → practice exams

Find the question types that keep coming back. Solve five of a type, then ask for five new ones, or a whole mock exam.

>Prompt
07 / Habit 2
// habit 2

Make it teach, not tell.

tip 02

Slow it down

"Don't give me the answer. Ask me one clarifying question and wait."

tip 04

Find your blind spots

"What do I assume is obvious here? Where is my thinking weak?"

tip 05

Hard topic → sim

If the text doesn't click, ask for a small game, simulation or visual. I'll build one live in the demos.

>Prompt
08 / Habit 3
// habit 3

Keep your judgment.

tip 06

Second model as fact-checker

One writes, another checks every claim against the original sources. No source? Don't use it.

tip 07

End the day with a reflection

Tell it what you studied and where you got stuck, and get tomorrow's plan back.

tip 08

Critic, not assistant

"Be critical." "Evaluate this."

It cuts down the people-pleasing.

>Prompt
09 / Taste
// the skill that doesn't get automated

AI can write.
You decide what's good.

  • Taste: what's good, what's interesting, what's outdated
  • You're not here just to know things. You're here to learn to judge them
  • Every new student should ask: how do I use AI best in my studies?
After Heikkinen and Yli-Teevahainen, HS Visio, 23.8.2026
>Prompt
Section 02 / Agents

A prompt answers.
An agent keeps score.

How agents work, and why they're the next step after good habits

>Prompt
10 / Definition
// prompt vs. agent

A chat waits for you.
An agent runs the workflow.

  • reads your course material, not the whole internet
  • uses tools: reads files, writes notes, runs code
  • decides the next step, like quizzing you on what you got wrong last time
It does all this without you steering every step.
Anthropic, Building Effective Agents: the model directs its own process and tool use
>Prompt
11 / Your job
// what's left for the student

You define three things.

goals

What should it achieve?

"Get me exam-ready for ECON-A1000 by 23 October."

limits

What must it never do?

"Never give me an answer before I've tried."

materials

What may it read?

"Only my slides and old exams. Cite the page."

After Kiirikki, HS Visio, 23.8.2026
>Prompt
12 / Tips → agent
// an agent is your habits on autopilot

Every tip becomes a part.

tip 01 · project→
shared store a folder of your materials, notes and progress
tip 03 · old exams→
/exam a specialist that writes practice exams
tip 02+04 · teach→
/quiz asks one question at a time and logs your mistakes
tip 06+08 · judge→
quality gate cite the page or cut the line
tip 07 · reflect→
/review reads your progress and plans today
>Prompt
13 / Memory
// files are memory

The whole agent is one folder.

$ tree econ-a1000/econ-a1000/ ├── CLAUDE.md ← goals + limits ├── materials/ ← slides, old exams · the only source of truth ├── plan.md ← every exam topic + status ├── notes/ ← one cheat sheet per topic ├── progress.md ← scores + mistake log └── .claude/skills/ ← map · learn · quiz · review

Close the laptop on Monday. Open it on Thursday. It remembers what you got wrong.

>Prompt
14 / Quality gates
// where study tools lie to you

A confident wrong answer
is worse than none.

failure · invents

Cite or cut.

Every line points to a page. Not in your material? It says so.

failure · leaks

Answers stay hidden.

One question at a time. No answer until you've tried.

failure · flatters

Critic mode on.

Ask it to evaluate, not to agree. Check it with a second model.

>Prompt
Demo 1
// student demo · ~7 min

A study assistant.

name · programme

  1. ?
    What does it do for you every week?
  2. ?
    Goals, limits, materials: what did you give it?
  3. ?
    Where did it go wrong, and what did you change?
>Prompt
Demo 2
// student demo · ~7 min

A tutor for one course.

name · course

  1. ?
    Why this course, and what was hard about it?
  2. ?
    How do you stop it from just giving you answers?
  3. ?
    Has it changed how you study, or your grade?
>Prompt
Live build
>
// tip 05, live

Hard topic →
simulation.

Demo 3, mine: from one of my own courses. About 8 minutes. Watch the prompts, not the code.

>Prompt
Live build / The prompts
// three prompts · all on the build page
01 · find it Here are my lecture slides. Which concept is hardest to understand from text alone? Give me 3 candidates and say why. 02 · build it Build an interactive simulation of ⟦concept⟧ from my slides. Use only formulas and definitions from the slides and cite the slide for each. Include sliders for the key variables, a live chart, a one-line explanation that updates as I move the sliders, and a "predict first" mode with 3 challenges where I have to guess before it shows the answer. 03 · check it Be critical. List every formula and claim in the simulation with its slide. Flag anything that isn't in my slides, then fix it.
>Prompt
Section 03 / Build
// build block · 40 min

Build it.

  1. 01
    Open the build page every prompt tonight is copy-paste from there
  2. 02
    Step 0: make a project for one course 5 minutes · everyone does this first
  3. 03
    Pick one build from the menu and finish it for your own course
>Prompt
15 / The menu
// pick one · free AI works unless marked
A · tip 03 · 15 min

Practice-exam generator

Old exams → recurring question types → a fresh mock exam with an answer key.

B · tip 05 · 15 min

Hard topic → simulation

What I just built live, for your own course.

C · tips 02+04 · 10 min

Tutor that asks first

It never answers straight away, and it finds your blind spots.

D · tip 06 · 10 min

Source checker

A second model checks every claim against your slides.

E · tip 07 · 10 min

Daily reflection planner

Five minutes at night, and tomorrow's plan is waiting.

F · all tips · 30 min · paid

Full study agent

Claude Code in a course folder: map, learn, quiz, review. It remembers.

// add-on for any build: tip 08, critic mode
>Prompt
16 / Show & tell
// 2 minutes each

Show us what
you built.

  1. 01
    Which course, which build?
  2. 02
    Where did it lie, leak or flatter?
  3. 03
    What did you change to fix it?
>Prompt
17 / Take this with you
// this period

Take this with you.

// context
one project per course

Never start a study session from an empty chat.

// teach
"ask me first"

If it answers too fast, you're not learning. You're reading.

// judge
cite or it didn't happen

No source, no trust. Your taste decides what's good.

>Prompt
18 / Next
// keep building

Coming up.

Sep 21
Build Night 3: Build a Personal Assistant Startup Sauna · 17:15
Sep 26–27
Prompt Marketing Hackathon
Sep 29
Workshop 6: Agents for Studies, again friends stuck on the waitlist? send them this one
>Prompt
19 / Thanks
>Prompt
prompt.aaltoes.com
// SOURCES
Ada-Sofia Kari: Superkäyttäjät antavat vinkit opiskelijoille, HS Visio, 23.8.2026 · Anthropic: Building Effective Agents, 2024
Roediger & Karpicke: Test-Enhanced Learning, Psychological Science, 2006 · Cepeda et al.: Distributed Practice in Verbal Recall Tasks, Psychological Bulletin, 2006
thanks for building.
>Prompt