The Attention Budget
Why more context often makes things worse. How to feed information strategically.
The Concept
AI has a limited context window — think of it as working memory. It can only "see" so much at once.
Here's the counterintuitive part: more isn't always better. More context often makes output worse because attention gets diluted.
The Rule:
Give the AI exactly what it needs to do the job. No more, no less.
The Desk Analogy
Cluttered Desk
Important things get buried under context that doesn't matter for the current task.
Result: Unfocused, generic output
Clean Desk
The AI sees exactly what matters. Work gets done efficiently and accurately.
Result: Focused, precise output
When you dump everything into the conversation hoping the AI will "figure it out," you're creating a cluttered desk. It has to sift through noise to find the signal.
Practical Tips
1. Front-load what matters
Put the most important information first. The AI pays more attention to what comes early.
Bad:
"Here's my entire project history... oh and I need a subject line for this email."
Good:
"Write a subject line for this sales follow-up email. Context: meeting yesterday, discussed pricing, they seemed interested."
2. Don't dump everything
More pages doesn't mean better output. Give context that's relevant to THIS task.
Ask yourself:
"If I were delegating this to a human, what would they actually need to know?"
3. Long conversations drift
After 10-15 back-and-forth messages, the AI starts losing track of what you're actually trying to do.
Solution:
Start a new conversation and restate what you need with fresh context.
4. Restating helps refocus
If output quality drops mid-conversation, summarize what you need again.
"Let me refocus: I need three subject line options for a cold outreach email to CTOs. Keep them under 50 characters."
Finding the Right Altitude
There's a sweet spot between too specific and too vague. Here's how to find it:
| Too Specific | Just Right | Too Vague |
|---|---|---|
| Micromanaging every detail, leaving no room for the AI to think | Clear intent with room to execute | "Help me with this" |
| Brittle — breaks if anything changes | Flexible, consistent results | Inconsistent results every time |
Example:
Too specific:
"Use exactly 47 words. Start with 'The'. Use the word 'innovative' in sentence 3..."
Just right:
"Write a 50-word product description. Professional tone. Focus on time savings."
Too vague:
"Make it sound good"
The Attention Budget Test
Take a prompt you've used recently. Try cutting it in half. Compare the results.
You'll often find:
- •The shorter version is clearer
- •The AI focuses on what actually matters
- •Output quality goes up, not down
Key Takeaway
The AI's attention is finite. More context ≠ better output. Give it exactly what it needs to do the job. No more, no less. Front-load what matters. Trim the rest.