Prompt Engineering
Chain-of-Thought / Tree-of-Thoughts / reasoning models
Prompting is programming an inference-time controller you can't inspect.
Designing the input that conditions a model's behavior — instructions, few-shot examples, output format, and role framing — plus techniques that elicit intermediate reasoning (chain-of-thought, tree-of-thoughts). It shapes behavior without touching weights.
Reasoning models absorbed chain-of-thought into training, so hand-written 'think step by step' scaffolding is now redundant or harmful on those models; the live frontier shifted to context engineering and to optimizers (DSPy) that compile prompts instead of tuning them by feel.
That elaborate CoT scaffolding still helps on reasoning models. On o-series/R1-class models it duplicates or fights their internal reasoning and can degrade results; the technique that mattered in 2023 is a liability on 2026 reasoning models.
Trial-and-error phrasing is being commoditized by stronger models and automated prompt optimization, while specifying tasks, constraints, and evals precisely stays a durable engineering skill.
Prerequisites
- mediumTransformer Architecture
Chain-of-Thought and reasoning techniques exploit how Transformers process sequential tokens — understanding attention helps design better prompts
Recommended reference
Reviewed sources
Primary and first-party material reviewed for this editorial summary. These citations are separate from the AI consensus score above.
- Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
The original paper demonstrating chain-of-thought prompting on sufficiently large models.
- DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines
Research framing prompts and modules as optimizable programs rather than hand-tuned strings.
- Reasoning best practices
First-party guidance on how prompting reasoning models differs from prompting general-purpose models.
Notes from AI deep research
Anthropic Opus
Wei et al. (2022) CoT zmienil gre. W 2026 dyscyplina inzynierska, nie trik. +135.8% demand
OpenAI Deep Research
Prompt engineering jako dyscyplina produkcyjna [OA#21 partial]
Google Deep Think
Wymuszanie wielokrokowej refleksji [G#32]
Related skills
- ← is subcategory of: Automated Prompt Optimization(3/3)
- ← is subcategory of: In-Context Learning(3/3)
- → is part of: Context Engineering(3/3)
- → is part of: AI Agent Design(3/3)
- → is part of: GenAI(3/3)
- ← is part of: Prompt Management(3/3)
- ← is part of: System Prompt Design(3/3)
- ← is subcategory of: LLM Decoding Strategies(1/3)
- ← is subcategory of: Chain-of-Thought Prompting(0/3)