Claude Code Engineer Details Agentic Coding in the Fable 5 Era
新智元 · wechat · 2026-07-04
Anthropic Claude Code core engineer Thariq points out that at Fable 5 intensity, the bottleneck has shifted from model capability to the user's ability to "clarify the unknown." He categorizes unknowns into four types: known knowns, known unknowns, unknown knowns, and unknown unknowns. "Unknown unknowns" are the most destructive, and the core competency of a top-tier agentic coder is reducing and planning for these unknowns.
Thariq provides a three-phase SOP (pre-, mid-, and post-implementation): pre-implementation involves "blind spot scanning," using HTML prototypes to quickly expose visual preferences, having Claude ask back key architectural questions, providing reference via source code instead of verbal descriptions, and writing an implementation plan with the most likely changes placed first; mid-implementation involves having Claude maintain an implementation-notes.md to record deviations; post-implementation involves packaging pitch docs and having Claude generate tests to gauge understanding depth.
The article emphasizes that writing prompts is a tightrope walk: being too specific causes Claude to stubbornly cling to instructions, while being too vague leads it to guess best practices on its own—both extremes lead to failure. Reducing unknowns and planning for them is the very essence of the craft of agentic coding.
More from coding & agent
- Alex Townsend posts 200 open problems in numerical linear algebra for humans and AI agents — IgorCarron · 2026-09-11
- Kimi K2.8 Preview rolls out: near-K3 coding performance, 1M context for all tiers — teortaxesTex · 2026-09-11
- Looking for a classifier of software engineering task shapes to pick models per task — StewartalsopIII · 2026-09-11
- Steal this idea: prompt-to-hardware where agents assemble custom devices — paraschopra · 2026-09-11
- Model Is the Least Interesting Part: A Guide to Six Core AI Architectures from RAG to Multi-Agent — goyalshaliniuk · 2026-09-11
- Non-coder builds layered memory architecture: 20k tokens tracks a year of agent conversations — matteoianni · 2026-09-11