Chain-of-Thought Prompting
Chain-of-thought prompting is a prompting technique that asks the model to reason step by step, improving accuracy on multi-step problems.
In plain English
"Let's think step by step." That is chain-of-thought prompting in one phrase. For tasks that need reasoning — math, multi-hop questions, plan decomposition — asking the model to show its work before committing to an answer boosts accuracy noticeably. The intermediate steps serve as scratchpad: the model explores, corrects itself, and lands on a better final answer than if it had leapt straight to the output.
Modern models have chain-of-thought partly baked in, especially reasoning-tuned variants that do long internal deliberation before responding. Even so, explicit "reason first, then answer" prompting still helps in agent contexts where the model needs to plan before acting. The trick is to separate the thinking from the output so downstream parsers can grab the answer cleanly.
Why it matters for Black Box
Black Box's CEO agent uses explicit chain-of-thought for planning turns: list candidate approaches, score them, pick one, then delegate. Specialists use it for anything with branching logic. The thinking is visible to the owner in the Action Feed, which is part of the product's transparency.
Examples
- Asking a model to explain its approach before computing a forecast.
- A planner writing a bullet list of steps before emitting the structured plan.
- An evaluator listing each rubric item's verdict before the overall pass/fail.