Curated by
More in AI Software Development
See all 33 →Khairallah AL-Awady on X: Anthropic engineer insights
x.com
Khairallah AL-Awady on X: Link Sharing
x.com
Sudo su on X: My Workflow with 4-6 Agents
x.com
More from Stacklist Team
See all stacks →Films Where Women Run the Show 👑
11 cards · Movies
What teas actually help with bloating?
6 cards · Food
Books Set in New York City 🗽
6 cards · Books
Using Prompts to Optimize Token Usage in AI Models
hoeem shares a two-prompt workflow strategy for reducing token waste in expensive AI coding tools like Claude Code and Codex by first planning in a cheaper model. The first prompt, attributed to Karpathy, establishes a detailed system persona emphasizing accuracy, intellectual rigor, independent reasoning, and explicit confidence levels.
Built for AI agentsACO · 1053 tokens
Summary
hoeem shares a two-prompt workflow strategy for reducing token waste in expensive AI coding tools like Claude Code and Codex by first planning in a cheaper model. The first prompt, attributed to Karpathy, establishes a detailed system persona emphasizing accuracy, intellectual rigor, independent reasoning, and explicit confidence levels.
Tags
prompt-engineering · token-optimization · claude-code · codex · ai-workflow · cost-efficiency · system-prompts
Key entities
hoeem (person, 0.9) · Andrej Karpathy (person, 0.85) · Claude Code (technology, 0.95) · Codex (technology, 0.9) · token optimization (concept, 0.95) · prompt engineering (concept, 0.9) · X (Twitter) (organization, 0.7)
Classification
snippet · language en · status final
Provenance
claude-opus-4-6 via @stacklist/mcp-server@2.0.0, confidence 0.85, 16 Jun 2026