Tool use, function calling, orchestration and the protocols that let models act rather than only answer.
The article discusses the rise of large language models (LLMs) and their growing influence on our decisions and actions. It explores the concept of agentic LLMs—LLMs that can act directly in the real world and interact with one another. The book aims to explain how these agentic LLMs function, their capabilities, and the implications for users.
The article argues that AI agents have surpassed human capacity in code production, prompting a shift in perspective from merely using AI to actively employing it. It highlights how this development has altered the author's view on AI’s role in software development. The piece suggests that embracing AI as a collaborator is now essential.
Agent4Spec is an autonomous multi‑agent system that uses large language models to continuously improve API specifications. It includes a validation pipeline that detects and corrects OpenAPI syntax errors, and a market‑analysis pipeline that researches competitors, finds functional gaps, and auto‑generates missing API features. In tests on two curated datasets, the best LLM achieved 100 % syntax error detection, over 95 % success in implementing corrections, and more than 50 % adoption of suggested endpoint improvements.
Large language model agents are vulnerable to memory‑poisoning attacks, and existing defenses rely on external modules that are memory‑inefficient and computationally heavy. The paper introduces DREAM, a training‑free, agent‑intrinsic memory repair framework that uses weak reactivation to destabilize malicious memories, anomaly detection on activation patterns, and selective suppression of harmful memories while reinforcing benign ones. Experiments show DREAM cuts backdoor attack success by over 95 % and improves runtime and token efficiency compared to A‑MemGuard, while maintaining high task success on poisoned multi‑agent systems.
Asana used GPT‑6 Astra in Codex to reduce the cost of its browser agent by 76 times and increase its speed by five times in tests. The improvement allows the company to provide customers with more capable models at a lower cost and faster performance.
MotionCoach is a multimodal agent designed to enhance video editing models for professional motion editing, such as sports and martial arts. It generates grounded prompts and reference clips through multi‑turn tool use and self‑verification, enabling precise motion guidance without retraining the base editing models. The system is trained on a curated dataset and evaluated on VPMEBench, achieving significant score improvements on benchmark datasets.
AgentFly is an agentic reinforcement learning framework that introduces a unified resource layer to manage heterogeneous environments—such as sandboxes, model services, and external APIs—through a single scheduling engine. The framework is organized into four layers: an agent layer that abstracts agents, tools, and rewards; a rollout layer that composes agent loops and computes rewards; a context layer that organizes rollouts and injects contextual information; and a low‑level resource layer that handles resource management. AgentFly offers prebuilt tools and environments, demonstrates successful agent training across multiple tasks and models, and provides the first controlled cross‑framework throughput comparison against other agentic RL frameworks.
Policy Learning with a Language Bottleneck (PLLB) is a framework that lets AI agents generate linguistic rules to describe high‑level strategies. PLLB alternates between rule generation, guided by language models, and policy updates that use these rules, even when a single rule cannot fully capture a complex policy. Experiments on five tasks—including a signaling game, maze navigation, image reconstruction, and robot grasp planning—show that PLLB agents learn more interpretable, generalizable behaviors and can share rules with humans for better coordination.
Project Greenhouse explores building fully open and sovereign agentic search models with modest computational resources. The authors present a competitive pointwise decoder-only reranker trained from scratch and fine-tuned on common datasets, avoiding reliance on third‑party open‑weight backbones. They demonstrate that most experiments can be run on a handful of GPUs and release data, code, configurations, and checkpoints for reproducibility.
SkillContrast is a training‑free selector that compares retrieved agent skills and keeps only the text that differs between them, preserving local context for a pretrained reranker. In experiments on 1,235 requests from SameCapRisk‑Bench, it improves clean‑hit rates by 54–72 over TF‑IDF query selection while using 51.1–58.8% fewer input tokens. The method shows that retaining differing text, rather than full skill bodies, is the key factor for better reranking performance.
The paper proposes that self‑evolving large‑language‑model agents can only be safely reviewed if their changes are confined to the runtime harness—instruction text, tool‑call logic, and primitive composition—while the model weights remain fixed. A dual‑loop engine with an admission gate records a hash‑chained diff before deployment, and simulation experiments show that this gate admits only safe changes, preventing error increases and maintaining false‑positive rates. When the gate is removed, many harmful changes slip through and missed flags rise, illustrating the gate’s protective role.
The paper introduces PF‑LSA, a minimalist algorithm for personalized federated linear stochastic approximation that blends each agent’s local update with a global average. It achieves both fast convergence to each agent’s personalized solution and a linear speedup in the number of agents when their learning problems are similar, without requiring prior knowledge of heterogeneity. The analysis decomposes error into consensus and disagreement components, showing rapid decay of consensus error and negligible disagreement in low‑heterogeneity regimes.
The study investigates whether large language models (LLMs) rely solely on their initial task context or adapt based on experimental feedback when selecting fine‑tuning configurations for neural operators. Experiments across partial differential equation families show that LLMs outperform random search and Bayesian optimisation, with early configurations already near the top of a random‑search pool and subsequent proposals changing in response to validation scores. Controlled interventions confirm that the LLM combines a task‑dependent prior with sensitivity to observed outcomes.
The paper proposes a new paradigm called Code-Only-as-Policy (COAP), treating the embodied world as a Turing machine where the robot and environment state are the tape and the policy is the code. COAP replaces traditional vision-language agents by using code to measure, track, and decide based on camera images and proprioception, enabling explicit state storage, fast execution, and extensibility across tasks. On RoboDojo’s 42 bimanual tasks, the COAP library achieved a 70.24% success rate without a model at test time.
The paper presents methods for black-box forensic analysis of conversational large language model (LLM) agents, focusing on two tasks: attribution—identifying the base model powering a chatbot—and fingerprinting—detecting identical system prompts across different endpoints. Attribution classifiers achieve 98% accuracy using only a few non‑adversarial conversation turns, while a cross‑encoder fingerprinting approach attains an AUC of 0.768 and an F1 of 0.703 on unseen prompts, improving to 0.943 AUC when aggregating 50 interactions. These techniques enable investigators to trace AI‑enabled scams back to their underlying models and link related scams through shared system prompts.
Opera is a verbal critic framework designed for long‑horizon coding agents, treating each correction as a persistent note that is followed until the problem is resolved. It uses periodic and event‑driven triggers to diagnose issues, audits feedback against visible evidence, and tracks subsequent actions to distinguish compliance from true resolution. In tests, Opera improves the resolve rate of non‑critic agents by up to 12.4, 15.0, and 8.9 percentage points on three benchmarks and provides on‑policy training data that boosts a model’s performance without requiring a critic at inference time.
Chronos is a test‑time framework that transforms historical pull requests into structured experience cards linked by a typed graph of code, developer intent, and organizational relations. It uses semantic search and weighted multi‑hop expansion to retrieve relevant cards, guiding both patch generation and selection by LLM‑based code agents. The approach improves resolution rates across multiple LLM backbones on SWE‑Bench and related benchmarks, and human evaluation shows a higher number of useful cards retrieved compared to flat semantic retrieval.
ViSkill is a visual-native skill learning framework that encodes successful interactions into composite visual skill cards for VLM agents. These skill cards guide inference and reward shaping, while distilled trajectories are reintegrated into the library, creating a closed feedback loop that reinforces both skill accumulation and policy improvement. Evaluated on Sokoban, FrozenLake, and PrimitiveSkill, ViSkill achieves a 0.89 success rate (0.91 with cold‑start), outperforming existing baselines and converging faster than standard PPO.
SkillGraph introduces a skill library framework that models reusable skills as nodes in a directed graph, with typed edges representing prerequisite, enhancement, and co-occurrence relationships. When presented with a new task, the system retrieves an ordered subgraph of relevant skills, guiding multi-step decision making. The graph is continuously refined through agent trajectories and reinforcement learning, enabling simultaneous improvement of the skill library and the agent policy, and achieving state‑of‑the‑art results on ALFWorld, WebShop, and several search‑augmented QA tasks.
SERL‑SQL introduces a selective execution‑grounded reinforcement learning framework for multi‑turn Text‑to‑SQL agents. It samples on‑policy interaction trajectories and employs a training‑only teacher to re‑score student actions using execution feedback, converting the teacher‑student likelihood gap into bounded, masked weights that reweight GRPO advantages only on SQL and tool‑action tokens. Experiments on BIRD, Spider, and cross‑domain benchmarks show competitive performance, achieving 76.56% execution accuracy on BIRD‑Dev and 89.92% on Spider‑Test, and the reward‑based selection strategy approaches the oracle Best‑of‑N upper bound.