Why it matters

These three are how an agent scales from "smart chatbot that can run a shell" to "specialist that reliably does your team's actual workflows." Skills let you encode institutional knowledge once and reuse it cheaply. MCP lets you connect the real systems where your data lives. Deterministic computation is what makes the output trustworthy, which, as Chapter 13 and 16 argue, is the whole ballgame once an agent is doing work that matters. An agent that confidently fabricates is worse than no agent; deterministic computation plus the anti-hallucination contract is the structural cure.