CodeCon ZA Presentation Structure

CodeCon ZA Presentation Structure

Topic: Nie prompt, ale kontext prináša výsledok Duration: 20 min + 10 min Q&A Audience: ~1000 (developers, managers, HR, CEOs) Speaker: Oliver Kriska


QUICK REFERENCE CARD

THE STORY IN ONE SENTENCE:
"You're giving AI little context and expecting great results -
here are 10 practical ways to fix that, starting today."

KEY QUOTES TO REMEMBER:
• "Giving little, expecting much" (the problem)
• "AI is like a colleague who knows nothing but can learn everything"
• "Iteration is not failure - it's refinement"
• "AI won't replace you - it helps you surpass yourself"

THE 10 ADVICES:
BEFORE: 1.WHY 2.Atomic 3.Examples 4.Exclude 5.Done
AFTER:  6.2-min rule 7.Explain why wrong 8.Start fresh 9.Opinion not validation 10.Junior test

THE HOOK: Double Punch (2 minutes)

Part 1: The Image Reveal (45 seconds)

Slide 1: Show a cropped/zoomed photo that looks scary:

  • Close-up of sharp teeth, dark background
  • Looks like a dangerous, aggressive animal
  • Audience feels tension

Say: "What do you see?" (pause 3 seconds, let audience assume the worst)

Slide 2: Zoom out - reveal full image:

  • It's a happy dog mid-yawn
  • Tail wagging, playful setting
  • Completely harmless

Say: "Same image. Without context, your brain saw danger. With context, it's just a sleepy puppy."

Visual suggestion: Search stock photos for "dog yawning close up teeth" - crop to show only teeth/mouth first, then reveal the full adorable dog.


Part 2: The Text Message Panic (45 seconds)

Slide 3: Show messages appearing one by one (animation):

Boss: We need to talk.

(pause 2 seconds)

Boss: Come to my office when you can.

(pause 2 seconds)

Boss: It's about your future here.

Say: "Recognize this feeling?" (let the dread sink in)

Slide 4: Final message appears:

Boss: We're promoting you! 🎉

(audience laughs/relieves)

Say: "Three messages without context - your heart stopped. One message with context - you celebrate."


Part 3: The Bridge to AI (30 seconds)

Slide 5: Simple text on dark background:

This is what you're doing to AI.
Every. Single. Day.

Say: "Your brain just made two wrong assumptions in 30 seconds because of missing context. AI does this with every prompt you send. The difference? AI doesn't ask follow-up questions. It just... guesses. And then you blame the AI."

Transition: "In the next 18 minutes, I'll show you how to fix this."


SECTION 1: THE PROBLEM (2.5 minutes)

Slide 1.1: The Universal Mistake

Title: "Giving little, expecting much"

Visual: Scale/balance image - tiny input on one side, massive expectation on other

Say: "When I started with AI, I was giving large tasks, wasn't specific, provided little context, and expected a lot. Sound familiar? This is the #1 reason people give up on AI."


Slide 1.2: Prompt vs Context

Title: "Prompt ≠ Context"

Visual: Simple diagram:

┌─────────────────────────────────┐
│           YOUR INPUT            │
├─────────────────────────────────┤
│  PROMPT = What you want         │
│  (task, question, instruction)  │
├─────────────────────────────────┤
│  CONTEXT = Everything else      │
│  (data, examples, constraints,  │
│   priorities, format, style)    │
└─────────────────────────────────┘

Say: "Most people focus on the perfect prompt and ignore context. That's like giving a junior developer a task title and expecting perfect production code."


Slide 1.3: The Mindset Shift

Title: Your Core Philosophy

Visual: Large quote on screen:

"AI is like a person who knows nothing about your project,
 but can learn everything to extreme depth -
 if you explain what you need."

Say: " AI already has the knowledge, you just need to guide it to YOUR specific needs. AI already knows how atoms split. It can generate a picture for your kid's birthday. The knowledge is there. But it knows nothing about YOUR project, YOUR problem, YOUR context. You don't teach it - you guide it. You explain what you're expecting. And that requires context."


Slide 1.4: Industry Confirmation

Visual: Andrej Karpathy photo + quote:

"Context engineering is the delicate art and science
 of filling the context window with just the right
 information for the next step."

 — Andrej Karpathy
   (ex-Tesla AI Director, OpenAI founding member)

Also mention: Tobi Lutke (Shopify CEO), Harrison Chase (LangChain CEO) use this term

Say: "This isn't just my opinion. The best minds in AI are moving from 'prompt engineering' to 'context engineering'."


SECTION 2: THE 10 ADVICES (10 minutes)

Format for each slide:

  • Advice title (large, memorable)
  • Tech example (green background or icon)
  • Non-tech example (blue background or icon)
  • ~1 minute per advice

PART A: BEFORE YOU PROMPT (5 advices, 5 min)

Slide intro for Part A:

PART A: Before You Prompt

These 5 advices help you PREPARE
better context before you hit enter.

Visual: Planning/preparation icon (clipboard, checklist)


Advice #1: Give the WHY, not just WHAT

Visual:

💻 TECH:
❌ "Fix this bug"
✅ "Fix this bug because users can't save forms.
    Here's the error log and the file where it occurs."

📝 NON-TECH:
❌ "Write me an email"
✅ "Write an email to decline a meeting politely.
    I'm overloaded this week and this meeting isn't priority."

Say: "AI doesn't know your priorities. Tell it WHY you need something, and it can make better decisions."


Advice #2: Break into atomic parts

Visual:

💻 TECH:
❌ "Build me an expense tracking app"
✅ "Create HTML table with 3 columns: amount, category, date.
    Users can add rows. Store in localStorage."

📝 NON-TECH:
❌ "Plan my vacation"
✅ "Find 3 hotels in Rome under 100€/night with breakfast.
    Must be within walking distance to Colosseum."

Say: "Big task = random output. Small task = precise output. Break it down like you would for a new team member."


Advice #3: One example beats 1000 words

Stop explaining, start showing

Visual:

💻 TECH:
❌ "Write tests like we do"
✅ "Here's our test file [attached]. Write similar tests
    for this new function, same structure and naming."

📝 NON-TECH:
❌ "Write in my style"
✅ "Here are 2 of my previous LinkedIn posts [attached].
    Match this tone - direct, no fluff, practical examples."

Say: "One example is worth 1000 words of description. Show AI what you want."


Advice #4: Say what to EXCLUDE

Visual:

💻 TECH:
❌ "Find authentication libraries"
✅ "Find auth libraries for Node.js. Exclude deprecated ones,
    must have TypeScript support and active maintenance (2024+)."

📝 NON-TECH:
❌ "Find me a car seat"
✅ "Find car seat for 120cm child. Exclude Amazon sellers.
    Only ADAC safety-tested. Removable washable cover."

Say: "AI will include everything unless you tell it not to. Define your boundaries."


Advice #5: Define what DONE looks like

Visual:

💻 TECH:
❌ "Optimize this query"
✅ "Optimize to under 2 seconds response time.
    Current EXPLAIN output attached. Can't change schema."

📝 NON-TECH:
❌ "Write product description"
✅ "Max 100 words. Include 3 key benefits.
    No superlatives. Tone: professional but friendly."

Say: "If you don't define success, AI will guess. And it usually guesses wrong."


PART B: DURING & AFTER (5 advices, 5 min)

Slide intro for Part B:

PART B: During & After

You sent the prompt. Now what?
These 5 advices help you ITERATE effectively.

Visual: Cycle/loop icon

Key message to say:

"Iteration is not failure. With good context, iteration is FASTER and results are BETTER. You're refining, not fixing."


Advice #6: The 2-Minute Rule

Visual: Large stopwatch icon

If AI takes more than 1-2 minutes for first useful output:

→ Task might be too big — break it down
→ Context might need adjustment
→ This is a SIGNAL, not a failure

Stop. Adjust. Continue.

Say: "When I see AI doing many operations I didn't ask for, or taking too long - I take it as a signal. Maybe the task is too big, maybe context needs adjustment. I stop, adjust, and continue. It's not failure - it's navigation."


Advice #7: Explain WHY it's wrong

Visual:

💻 TECH:
❌ "This is wrong, try again"
✅ "This is wrong because we use PostgreSQL, not MySQL.
    The syntax for UPSERT is different. Here's the correct format."

📝 NON-TECH:
❌ "The tone is off"
✅ "The tone is too formal. I need casual language
    because this is for Instagram, not a business report."

Say: "AI can't learn from 'wrong'. It learns from 'wrong because X'. Give it the lesson."


Advice #8: Start fresh when context is contaminated

Visual: Trash bin icon + "New Chat" button

Signs of contaminated context:

• AI keeps suggesting solutions you already rejected
• It includes things you explicitly excluded
• Corrections don't stick after 2-3 attempts

Solution: New session with clean, correct context from start

Say: "Bad output stays in context and misleads future responses. Sometimes it's faster to start fresh than to fix a broken conversation."


Advice #9: Ask for opinion, not validation

Visual:

❌ "Is my code good?"      → AI will praise you
❌ "Is my plan good?"      → AI will agree with you

✅ "How would YOU solve this problem?"
✅ "What are 3 risks in this approach?"
✅ "What am I missing?"

Say: "Current AI models are tuned to be agreeable. If you want real feedback, ask for alternatives and risks - not validation."


Advice #10: The Junior Developer Test

Visual: Simple illustration of person with question mark

Before every prompt, ask yourself:

┌─────────────────────────────────────────┐
│  "Could a colleague who started         │
│   yesterday complete this task          │
│   with my description?"                 │
├─────────────────────────────────────────┤
│  If NO  → Add more context              │
│  If YES → AI can handle it too          │
└─────────────────────────────────────────┘

Say: "This is my universal test. Works for code, articles, emails, anything. If a new person couldn't do it with your instructions, neither can AI."


SECTION 3: WHY THIS MATTERS (2 minutes)

Slide 3.1: Scientific Proof

Title: "More context ≠ Better results"

Visual: Graph showing performance degradation (if available)

Citation box:

Chroma Research (July 2025): "Context Rot" study

• Tested 18 leading LLMs (GPT-4, Claude, Gemini...)
• Performance DEGRADES as input tokens increase
• Even on simple tasks
• Different models fail differently:
  - Claude tends to abstain when unsure
  - GPT tends to hallucinate confidently

Say: "This isn't intuition - it's science. Throwing more data at AI makes it worse. You need the RIGHT context, not MORE context."


Slide 3.2: Senior vs Junior

Visual: Two people icons - "Junior" and "Senior"

What's the difference between junior and senior?

Same skills. Same tools. Different CONTEXT.

Senior knows:
• What to include
• What to exclude
• What worked before
• What failed before

You can simulate years of experience
by providing proper context.

Say: "A senior developer isn't smarter - they have better context from experience. You can give AI that same context explicitly."


SECTION 4: CLOSING (2.5 minutes)

Slide 4.1: Three Takeaways

Visual: Three large icons with text

1. BREAK IT DOWN
   Atomic tasks, not massive requests

2. CONTEXT > PROMPT
   What you provide matters more than how you ask

3. TEST IT
   "Would a new colleague understand this?"

Say: "If you remember nothing else - remember these three things."


Slide 4.2: Final Quote

Visual: Large text, dark background, centered



"AI won't replace your work.
 It will help you surpass yourself."


Say: "Copywriters fear AI will take their jobs. Developers fear the same. But AI isn't here to replace you - it's here to help you do things you couldn't do alone. The key is learning to work WITH it. And that starts with context."

(pause for effect)


Slide 4.3: Call to Action

Visual: Your contact info + QR code

Start today:

1. Take your next AI prompt
2. Add WHY you need it
3. Add one example of good output
4. See the difference

Say: "Thank you. I'm happy to take questions."


VISUAL DESIGN GUIDE

Color Scheme

  • Background: Dark (navy #1a1a2e or charcoal #2d2d2d)
  • Text: White primary, light gray secondary
  • Accent - Good: Green (#4ade80)
  • Accent - Bad: Red (#f87171)
  • Accent - Highlight: Purple or blue (#8b5cf6)

Typography

  • Titles: 48-60pt, bold
  • Body: 32-36pt, regular
  • Code/Examples: 24-28pt, monospace
  • Minimum readable from back: 24pt

Icons Style

  • Simple, flat design
  • Single color (white or accent)
  • No 3D effects, no gradients
  • Consistent style throughout

Slide Layout Template

┌─────────────────────────────────────┐
│ ADVICE TITLE                    #X  │
├─────────────────────────────────────┤
│                                     │
│  💻 TECH                            │
│  ┌─────────────────────────────┐   │
│  │ ❌ Bad example              │   │
│  │ ✅ Good example             │   │
│  └─────────────────────────────┘   │
│                                     │
│  📝 NON-TECH                        │
│  ┌─────────────────────────────┐   │
│  │ ❌ Bad example              │   │
│  │ ✅ Good example             │   │
│  └─────────────────────────────┘   │
│                                     │
└─────────────────────────────────────┘

Animation Guidelines

  • Preferred: Simple fade in/out
  • For examples: Show ❌ first, then reveal ✅
  • For hook: Messages appear one by one
  • Avoid: Flying, spinning, bouncing, zooming

Article Details

Category
context engineering new codecon materials
Published
November 26, 2025
Length
2,220 words
14,101 characters
~9 pages
Status
Draft Preview

More from context engineering new codecon materials

Stop Prompting. Start Briefing.

# Stop Prompting. Start Briefing. **Author:** Tomas Cupr **Date:** August 10, 2025 > "Thinking" ChatGPT class models are landing in more hands. Results still vary. The difference is not magic, it is...

Read article

Effective context engineering for AI agents

# Effective context engineering for AI agents **Source:** Anthropic Engineering Blog **URL:** https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents --- After a few years...

Read article