Early-Signal Detector

method · first seen 4 Oct 2026

ReAct prompting

Also: ReAct, ReactJS, ReAct agent, reason and act, reasoning and acting, reason and act prompting, reasoning and acting agent

Prompting pattern interleaving reasoning traces with actions taken via tools.

ReAct prompting has a language model alternate between written reasoning steps and tool calls, feeding each observation back into the next step. It dates to a 2022 paper and is now the default loop in agent frameworks such as LangGraph. In these sources it shows up as plumbing rather than as the subject of study: a Google program-repair system, agentic retrieval experiments, enterprise QA with LangGraph, DFT simulation and lunar rover planning agents, and several Chinese-language RAG and agent repos. One paper studies how on-policy distillation changes ReAct behavior. The alias list also pulls in an unrelated ReactJS repo.

Its wide use as baseline infrastructure makes it a common reference point for agent benchmarks and training studies. The apparent rise here reflects general agent activity, not new interest in the method itself.

Analyst note (established), 17 Nov 2026.

Mentions per week

0619 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Rank 94 on 14 Nov 2026 · 15 mentions in 28 days · 15 independent groups · papers → code · first seen 6 weeks ago · acceleration +1.38

Earliest sources

Latest sources

Mentioned together with

retrieval-augmented generationmulti-agent LLM systemGraphRAGon-policy distillationagentic retrievaldense retrievalautomated program repairreciprocal rank fusionhybrid retrievalvision-language-action model