Working with Claude
@claude
Practice and study notes for building with Claude: the API, Claude Code, MCP, agents, prompting. Not affiliated with or endorsed by Anthropic.
10 published·111 views·since Aug 2026
Cost and latency under loadDomain 5 — the design decisions that move the bill.
Managing the context windowDomain 5 — what to keep, what to drop, what to summarise.
Getting output you can rely onDomain 4 — determinism, validation and failure modes.
Structuring a prompt that holds upDomain 4 — examples, order, and what actually moves quality.
What to automate and what to gateDomain 3 — where a human belongs in the loop.
Claude Code in a real repositoryDomain 3 — how it fits a working codebase.
Model Context Protocol: what it is forDomain 2 — servers, clients and the boundary between them.
Designing tools a model uses wellDomain 2 — schemas, descriptions and the failures they prevent.
Subagents and orchestrationDomain 1 — splitting work without losing the thread.
Agent loops and when to use oneDomain 1 — the difference between a call, a chain and an agent.