Inside Anthropic's Architect Foundations exam: a candidate's recap of agents, MCP and orchestration questions
SlayerC20 · reddit · 2026-09-12
A candidate who just took Anthropic's Architect Foundations certification exam shared a detailed recap, warning it's far denser than a "skim the docs" cert — it tests whether you actually understand how agentic systems work, not just Claude Code usage.
Topics covered (from memory):
- Agents/tools: structured outputs (nullable vs optional), writing tool descriptions and few-shot examples, chaining tool calls, retry vs surfacing errors, MCP structured messages and error responses, where MCP context loads in a session
- Orchestration: subagents, context flooding, resuming and forking sessions, plan→execute→review vs direct execution vs interview-style, and the tool loop (tooluse, toolresult, endturn)
- Config/control: CLAUDE.md vs PreToolUse hooks vs settings. permissions, long-session context management, prompt-based guardrails vs deterministic safety, grep vs glob vs read
Study resources mentioned: a personal Anki deck with spaced repetition, an X thread breaking down exam concepts, and Clouding Academy practice exams. His verdict: it's a decent exam that forces you to think about how things work under the hood rather than prompt-engineering vibes.
More from coding & agent
- Building a buyer agent that negotiates 8 seller agents: what broke and what fixed it — mv_soura · 2026-10-31
- PostTrainBench traces hit 104k monthly downloads — author warns it's likely reward hacking — maksym_andr · 2026-10-31
- Building games with AI? Set up a 'tech art' test scene to tune your look fast — chongdashu · 2026-10-31
- Sanity open-sources pi-mcp-events: streaming MCP Events client for the Pi agent harness — knutmelvaer · 2026-10-31
- Skydive opens to all: watch your AI coworkers work in a 3D office — Scobleizer · 2026-10-31
- LangChain SF meetup: decision models and a third path for agent evals — LangChain · 2026-10-31