Library › Book 1 › Token Economics -- Why Architecture Choices Matter

Token Economics -- Why Architecture Choices Matter

AI API SQL Playwright Security Performance Mobile Accessibility CI/CD Git Linux Agile Strategy Career

4.1🔒The Context Window Is a Shared ResourceEvery AI model has a context window -- its working memory. The system prompt, conversation history, file contents, tool definitions, and…166 words

4.2🔒MCP Token Cost: The Tool Schema TaxEvery MCP tool exposes a JSON schema loaded into every API request:70 words

4.3🔒Skill Token Cost: The Markdown InjectionA browser-automation SKILL.md -- the Playwright CLI's generated skill, or Vibium's vibe-check before it -- is roughly 100 lines, about…65 words

4.4🔒Naive Arithmetic: A Labeled IllustrationBefore the real measurements, the naive per-step arithmetic is a useful intuition pump for WHERE the costs live. Treat it as an…137 words

4.5🔒What Measurement Actually ShowsMicrosoft benchmarked the same coding-agent task through both transports:255 words

4.6🔒What You Do with the Saved ContextRun the measured numbers against the ~200K working budget. A browser-heavy task consuming ~114K tokens via MCP leaves well under half the…109 words

4.7🔒Dollar Cost at Scale: An IllustrationToken prices change often, so treat the rate as a placeholder: assume an illustrative $15 per million input tokens and substitute your…113 words

4.8🔒When MCP's Token Cost Is Justified1. Structured page understanding. For accessibility auditing, the accessibility tree IS the data you need. 2. Sandboxed agents. The…104 words

4.9🔒Key Takeaways- The context window is a shared, finite resource -- browser automation competes with reasoning - MCP tool schemas cost ~3,000-12,500…114 words

4.10🔒Q&ASelf-Assessment Quiz1. What did Microsoft's benchmark measure for the same task via CLI skill vs. MCP? 2. Why is naive per-step arithmetic misleading, and in…177 words

4.11🔒Exercises[Beginner] Exercise 4.1: Compute the naive token cost of a 10-step test both ways (MCP: ~10,200 tokens/step; skill: ~130 tokens/step after…129 words

4.12🔒Career Translation- Conducted a token economics analysis comparing agent skill and MCP architectures using measured benchmark data (~27K vs ~114K tokens per…312 words

4.13🔒Q&AInterview Depth CheckPrompt: The naive arithmetic says CLI skills are 50x+ cheaper than MCP; Microsoft's benchmark says ~4x. Explain the gap. Which number do…402 words