Context Engineering - Making Every Token Count

Addy Osmani

September 09, 2025

Tweet Share

Transcript

1. CONTEXT ENGINEERING ADDY OSMANI MAKING EVERY TOKEN COUNT

2. Based on illustration by dexhorthy

3. WHAT ARE TOKENS? https://platform.openai.com/tokenizer

4. WHAT’S A CONTEXT WINDOW?

5. CONTEXT WINDOWS ARE LIKE LIMITED RAM Curation of what fi

   ts into RAM is analogous to “context engineering” ANALOGY

6. PROMPT ENGINEERING Clear instructions for models so they can accomplish

   a task https://www.youtube.com/watch?v=ysPbXH0LpIE PRELUDE

7. PROMPT ENGINEERING FAILURES STEM FROM CONTEXT MISMANAGEMENT https://www.instagram.com/p/DMc4ujyTuk6/?img_index=6

8. CONTEXT ENGINEERING MEANS PROVIDING AN AI WITH ALL THE INFORMATION

   AND TOOLS IT NEEDS TO SUCCESSFULLY COMPLETE A TASK – NOT JUST A CLEVERLY WORDED PROMPT. DEFINITION

9. ADD CONTEXT CONTEXT WINDOW USE CONTEXT IN EDITORS & IDES

10. VISUAL CONTEXT IS PRETTY POWERFUL

11. https://www.youtube.com/watch?v=ysPbXH0LpIE

12. https://context.addy.ie

13. None
14. None
15. None
16. None
17. None
18. None
19. None
20. None
21. None
22. None

23. GOOD CONTEXT IS HARD. PROBLEM

24. AI CODING PERFORMANCE OFTEN DIPS WHEN CONTEXT WINDOWS EXCEED ~50%

    FULLNESS PROBLEM

25. TOO LITTLE CONTEXT: VAGUE OR HALLUCINATED RESPONSES. TOO MUCH CONTEXT:

    DISTRACTED, UNABLE TO FIND RELEVANT INFO, OVER-INDEXING ON PATTERNS. BAD CONTEXT: POISONING, TRUSTING INCORRECT STATEMENTS OVER TRAINING.

26. TOO LITTLE CONTEXT: VAGUE OR HALLUCINATED RESPONSES. TOO MUCH CONTEXT:

    DISTRACTED, UNABLE TO FIND RELEVANT INFO, OVER-INDEXING ON PATTERNS. BAD CONTEXT: POISONING, TRUSTING INCORRECT STATEMENTS OVER TRAINING.

27. TOO LITTLE CONTEXT: VAGUE OR HALLUCINATED RESPONSES. TOO MUCH CONTEXT:

    DISTRACTED, UNABLE TO FIND RELEVANT INFO, OVER-INDEXING ON PATTERNS. BAD CONTEXT: POISONING, TRUSTING INCORRECT STATEMENTS OVER TRAINING.

28. PATTERNS FOR CONTEXT MANAGEMENT https://blog.langchain.com/context-engineering-for-agents/

29. WRITE CONTEXT • Long-term memories (across sessions) • Scratchpad (within

    session) • State (within sessions) @aifolksorg

30. SELECT CONTEXT • Retrieve relevant tools • Retrieve from scratchpad

    • Retrieve long-term memory • Retrieve relevant knowledge. @aifolksorg

31. COMPRESS CONTEXT • Summarize context to retain relevant tokens •

    Trim to remove irrelevant tokens @aifolksorg

32. ISOLATE CONTEXT • Partition context in state • Hold in

    environment/sandbox • Partition across multi-agents @aifolksorg

33. AUTOMATIC CONTEXT OPTIMIZATION CURSOR CLINE

34. HOW TO FIX YOUR CONTEXT https://www.dbreunig.com/2025/06/26/how-to- fi x-your-context.html

35. None

36. •Be precise: Vague requests lead to vague answers. The more

    specific you are, the better your results will be. •Provide relevant code: Share the specific files, folders, or code snippets that are central to your request. •Include design documents: Paste or attach sections from relevant design docs to give the AI the bigger picture. •Share full error logs: For debugging, always provide the complete error message and any relevant logs or stack traces. •Show database schemas: When working with databases, a screenshot of the schema helps the AI generate accurate code for data interaction. •Use PR feedback: Comments from a pull request make for context-rich prompts. •Give examples: Show an example of what you want the final output to look like. •State your constraints: Clearly list any requirements, such as libraries to use, patterns to follow, or things to avoid. CONTEXT ENGINEERING FOR AI CODERS

37. CONTEXT IT’S NOT THE PROMPT, IT’S THE ADDYOSMANI.COM