# Agentic Reasoning

Canonical URL: https://menfem.com/kb/harnesses/concepts/agentic-reasoning
Knowledge base topic: [Harnesses & Agent Systems](https://menfem.com/kb/harnesses)
Frontier status: active
Tags: paradigm, agents, reasoning

---

Agentic reasoning represents a paradigm shift in how we frame large language models — not as static question-answering systems, but as autonomous agents that plan, act, and learn through continual interaction with their environment. This reframing moves LLMs from passive tools to active participants capable of multi-step problem solving.

Wei et al. propose a three-layer framework that organizes the field: **foundational agentic reasoning** (single-agent capabilities like planning, tool use, and search in stable environments), **self-evolving agentic reasoning** (agents that refine capabilities through feedback, memory, and adaptation), and **collective multi-agent reasoning** (intelligence extended to collaborative multi-agent settings).

A critical distinction runs across all three layers: **in-context reasoning** (test-time interaction without weight changes) versus **post-training reasoning** (reinforcement learning optimization that updates model parameters). Production systems increasingly combine both approaches, using in-context reasoning for flexibility and post-training for robust capability internalization.

## Key Claims

- **Agentic reasoning is a paradigm shift for LLMs** — Moves models from static QA to autonomous planning, acting, and learning through interaction. *Evidence: strong* ([Agentic Reasoning for LLMs](../../raw/agentic-reasoning-for-llms.md))
- **Three-layer framework captures the field** — Foundational → self-evolving → multi-agent, with in-context vs. post-training as an orthogonal dimension. *Evidence: strong* ([Agentic Reasoning for LLMs](../../raw/agentic-reasoning-for-llms.md))
- **~60 benchmarks exist across 8 domains** — Evaluation landscape spans general reasoning, math, code, factual grounding, multimodal, and interactive tasks, developed 2019-2025. *Evidence: strong* ([From LLM Reasoning to Autonomous Agents](../../raw/llm-reasoning-to-autonomous-agents.md))
- **Production systems combine all three tool-use paradigms** — Prompting, supervised fine-tuning, and RL are complementary, not competing. *Evidence: strong* ([Agentic Tool Use in LLMs](../../raw/agentic-tool-use-in-llms.md))
- **Gap exists between benchmark and real-world performance** — Agent capabilities measured on benchmarks don't fully transfer to deployment. *Evidence: moderate* ([From LLM Reasoning to Autonomous Agents](../../raw/llm-reasoning-to-autonomous-agents.md))
- **VLA models are the physical instantiation of agentic reasoning** — Vision-Language-Action models unify perception, language understanding, and action generation, extending agentic reasoning from digital tool use to embodied robotic manipulation. *Evidence: strong* ([Efficient VLA Survey](../../raw/efficient-vla-models-survey.md), [VLM-VLA Robotic Manipulation Survey](../../raw/vlm-vla-robotic-manipulation-survey.md))
- **Safety is a critical open problem with 6 documented failure modes** — Reward hacking, sycophancy, annotator drift, alignment mirages, rare-event blindness, and optimization overhang represent systematic patterns of misalignment in agentic systems. *Evidence: moderate* ([AI Safety, Alignment, and Interpretability in 2026](../../raw/ai-safety-alignment-interpretability-2026.md))
- **Memory is the key infrastructure for self-evolving agents** — The write-manage-read loop with five mechanism families enables agents to persist knowledge across sessions, directly supporting the self-evolving layer of the three-layer framework. *Evidence: strong* ([Memory for Autonomous LLM Agents](../../raw/agent-memory-mechanisms-survey.md))
- **Planning failures are representational, not reasoning-bound** — Model-First Reasoning (MFR) proposes that LLMs should construct explicit problem models (entities, state, actions, constraints) *before* generating solutions. Ablation shows the modeling phase is load-bearing; CoT and ReAct baselines are outperformed across medical scheduling, route planning, resource allocation, logic puzzles, and procedural synthesis. This is a structural argument: failures attributed to "weak reasoning" often reflect implicit state tracking breaking under constraint complexity. *Evidence: strong* ([Model-First Reasoning LLM Agents](../../raw/model-first-reasoning-llm-agents.md))

## Benchmarks & Data

- 60 benchmarks taxonomized across 8 evaluation domains (2019-2025) ([Ferrag et al.](../../raw/llm-reasoning-to-autonomous-agents.md))
- Real-world applications documented across 11 sectors ([Ferrag et al.](../../raw/llm-reasoning-to-autonomous-agents.md))
- Evaluation matured from function-call metrics to holistic interactive benchmarks like WebArena and OSWorld ([Hu et al.](../../raw/agentic-tool-use-in-llms.md))

## Open Questions

- How to achieve robust long-horizon interaction (multi-step plans that span hours/days)?
- How to govern multi-agent systems — alignment, safety, accountability?
- Can agentic reasoning extend effectively to multimodal settings (vision, audio, physical)?
- How to personalize agent behavior while maintaining safety guarantees?

## Related Concepts

- [LLM Tool Use](./llm-tool-use.md) — The mechanism that operationalizes agentic action
- [Multi-Agent Systems](./multi-agent-systems.md) — The third layer of the agentic reasoning framework
- [Chain-of-Thought Reasoning](./chain-of-thought-reasoning.md) — Core reasoning technique within agentic systems
- [Reinforcement Learning for Agents](./reinforcement-learning-for-agents.md) — Post-training paradigm for optimizing agent behavior
- [Agent Evaluation Benchmarks](./agent-evaluation-benchmarks.md) — How agentic capabilities are measured
- [Vision-Language-Action Models](./vision-language-action-models.md) — Physical embodiment of agentic reasoning in robotic systems
- [Agent Safety & Alignment](./agent-safety-alignment.md) — Safety constraints and failure modes for autonomous agents
- [Agent Memory Architectures](./agent-memory-architectures.md) — Infrastructure for self-evolving agents (layer 2)

## Backlinks

*Pages that reference this concept:*
- [Google DeepMind](../entities/google-deepmind.md)
- [AlphaEvolve](../entities/alphaevolve.md)
- [Vision-Language-Action Models](./vision-language-action-models.md)
- [Agent Safety & Alignment](./agent-safety-alignment.md)
- [Agent Memory Architectures](./agent-memory-architectures.md)

## Changelog

- **2026-04-15** — Added Model-First Reasoning (MFR) claim: planning failures reframed as representational deficiencies rather than reasoning weaknesses. Paper: 2512.14474 (Rana & Kumar, Dec 2025).

## Sources

- agentic-reasoning-for-llms
- llm-reasoning-to-autonomous-agents
- agentic-tool-use-in-llms
- efficient-vla-models-survey
- vlm-vla-robotic-manipulation-survey
- model-first-reasoning-llm-agents

---

Cite as: MenFem Knowledge Base — https://menfem.com/kb/harnesses/concepts/agentic-reasoning