Train AI to shut up and keep up.

This newsletter: Find where AI fails you and show you the workarounds I’ve found actually work. For me, at least.

Format: As spartan as possible. The prompt at the top, chat at the back.

Why listen to me? I’ve spent 3 years trying to charm AI into doing what I want. 90% of it didn’t work. I wasted a lot of time. You don’t have to.

This week: The most useful hack I’ve found is to get AI to shut the hell up.

COMPRESSION MODE is designed to help you move fast, strip out distractions, filter results by a governing question.

The prompt: Paste this into any chat and run, baby, run. If you want the system prompts, give me a shout.

INTELLIGENT COMPRESSION MODE – APPLY TO ALL MY REQUESTS IN THIS SESSION: For every request I make in this conversation, 

Step 1: Extract the governing question – What decision am I actually making? (state in one sentence). 

Step 2: Generate + score silently – Create options internally, score each 1–10 against the governing question. 

Step 3: Give me top 3 only: 

FORMAT FOR SCAN-READING:
- **Bold the label** (1-4 syllables)
- 30 words: why this scores highest against governing question
- *Trade-off: what I gain vs what I sacrifice*
- One blank line between options

Lead with what's surprising about each option, not generic description.

Confirm by responding:
"Compression mode active. Your options:
• 'more' → see options 4-6 (still compressed)
• 'show all' → full unfiltered output for that request
• 'expand [1/2/3]' → full depth on one option
• 'compression off' → stop compression for rest of session"

Discussion

AI isn’t built for my brain.

Your brain holds 3-5 meaningful chunks at once. When working memory is overwhelmed, you stop making decisions and start managing information.

Processing feels like progress, but it erodes confidence in the decisions made. Before long we’re leaning on AI for affirmation, and we all know what a sycophantic hype man AI can be (I’ll go into that another time).

Compression mode cuts the results down to brain-friendly chunks. As a creative, it helps me move quickly without clogging up my mind.

But how AI summarises is a problem (again, something for another week and a different hack); identifying the governing questions and using them as the filtering reference stops it applying equal weight to all options. Telling you why it filtered to those options helps decide, rather than just process.

Nice theory, but in practice…

I tracked some typical creative brief-to-ideas sessions, before/after using compression constraints:

SINGLE REQUEST:

• 2,000 words to evaluate: ~16 minutes (reading + processing + deciding)
• 90 words compressed: ~3 minutes

 

FULL CREATIVE SESSION:

• 15,000-20,000 words across session: 90-120 minutes of reading/evaluating
• Add needless side paths and time to regain focus
• Total cognitive cost: 2.5-3 hours
• Same session with Compression Mode: under 500 words → ~20-30 minutes

Is that it? Yeah!

I've got other hacks that combine with this, but compression alone was a game-changer. I haven’t tried it in every model but even this simple in-chat prompt should change how your AI behaves. Try it and let me know.

 

One more thing: make your deadline AI’s deadline.

Before you start any AI session, tell it what you're trying to achieve and when.

I need three campaign concepts by 2pm. 
Map out a process to get there covering {strategy, 
insights, differentiating details, outside-in perspective etc. - adapt these steps to what makes sense for your brief}. 
If I drift from the goal, flag it and bring 
me back. Keep me moving.

Sure, you’ll drift, but it’ll try to keep you on track. It works on its own; even better with Compression Mode running.