Mortdecai
25e64c1f06
Replaced specific IPs, CT numbers, hostnames, and domains with generic examples suitable for public sharing. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
1.9 KiB
1.9 KiB
Why Use Shorthand with Claude?
The Problem
Every token costs money and time. A typical conversational prompt uses 2-3x more tokens than necessary — articles, filler words, politeness markers, and redundant context that Claude already knows.
The Math
| Style | Tokens/msg | 100 msgs/day | Monthly cost (Opus) |
|---|---|---|---|
| Conversational | ~80 | 8,000 | baseline |
| Shorthand | ~25 | 2,500 | ~70% less |
Output tokens cost 5x input on most Claude models — but shorter, denser prompts also produce shorter, denser responses. The savings compound.
What You're NOT Losing
- Accuracy — Claude parses structured input better than prose. Ambiguity drops.
- Nuance — Operators like
?:!express conditionals and negation precisely. - Context — CLAUDE.md aliases (short names for servers, services, pools) carry full meaning in 1 token.
What You ARE Gaining
- Speed — Less typing, faster responses, shorter round-trips
- Clarity — Structured specs force you to think about what you actually want
- Cost — 50-70% token reduction on input, cascading savings on output
- Signal density — Every token carries meaning, nothing wasted
When to Use Full Prose
- Teaching Claude something new (novel context it can't infer)
- Complex reasoning where relationships between ideas matter
- When you'd write prose to a human colleague too
The Mental Model
Think of it like SQL vs. English:
-- English:
"Could you please look at all the users in the database who signed
up after January 1st and haven't logged in for 30 days, and give
me their email addresses?"
-- SQL:
SELECT email FROM users WHERE created > '2025-01-01' AND last_login < NOW() - 30d
Both say the same thing. One is 40 tokens. The other is 15. Claude understands both equally well.
Shorthand is your "SQL for Claude."