✏️ Prompt Engineering
Prompting patterns — few-shot, chain-of-thought, ReAct, structured output, prompt management at scale.
The workflow
flowchart LR
A[Task definition] --> B[System prompt<br/>role + rules]
B --> C{Pattern}
C -->|Few-shot| D[Curated examples]
C -->|CoT| E[Step-by-step]
C -->|ReAct| F[Thought + action loop]
C -->|Structured| G[JSON / XML schema]
D --> H[Iterate on traces<br/>+ regressions]
E --> H
F --> H
G --> H
Prompt engineering is engineering: version, test, regression-gate. The clever wording usually doesn't matter.
Key takeaways
Videos (12)
Context Is the New Code — Patrick Debois, Tessl
Context artifacts (prompts, skills, agent.md) need the same generation, testing, observability, and versioning rigor we apply to code.
Building with Anthropic Claude: Prompt Workshop with Zack Witten
Iterative prompt engineering with structured XML context and a test-case eval harness in the Anthropic Console produces measurably better Claude outputs than vibe-based editing.
Prompt Engineering is Dead — Nir Gazit, Traceloop
Replace manual prompt fiddling with an evaluator-in-the-loop optimizer agent that rewrites prompts until the judge score improves.
The Unreasonable Effectiveness of Prompt Learning – Aparna Dhinakaran, Arize
For most agent teams, eval-driven prompt learning with natural-language feedback beats RL on data efficiency and engineering cost.
Build a Prompt Learning Loop - SallyAnn DeLucia & Fuad Ali, Arize
Optimize prompts by feeding the LLM rich English feedback about *why* each failure happened, not just scalar scores or output diffs.
10x Development: LLMs For the working Programmer - Manuel Odendahl
Become a 10x engineer by treating the LLM as a translation/word-simulation engine and using concrete habits like regeneration, response-editing, transcript summarization and ridiculous domain framing.
Principles for Prompt Engineering - Karina Nguyen (Claude Instant @ Anthropic)
Prompting Claude well means treating it like a person, structuring inputs with consistent XML tags, and iterating like creative writing.
Optimizing LLMs in Insurance with DSPy: Jeronim Morina
Treat LLM apps as compilable programs: define metrics on real production data, then let DSPy optimize prompts rather than hand-tweaking templates.
The Model Isn't Wrong—You're Just Bad at Prompting
Prompt engineering is still the cheapest lever — but reasoning models invert the old rules, preferring minimal prompts with encouragement to think longer.
The Coherence Trap: Why LLMs Feel Smart (But Aren't Thinking) - Travis Frisinger
LLMs feel intelligent because of coherence, not understanding — engineers should design and evaluate around that illusion rather than be fooled by it.
Prompt Engineering Tactics: Dan Cleary
Three cheap, research-backed prompt edits—multi-persona, 'according to', and emotional stimuli—measurably lift output quality with no code change.
Your LLM Ran Out of Knowledge — Now What?
For low-knowledge domains, encode expert heuristics as a rule library and route prompts to the right rule set before reasoning — wisdom rules in, reasoning model on top.