AI agents advance with new frameworks for real-world tasks and benchmarks · 9 sources tracked
ByPulseAugur Editorial·[111 sources]·
New research explores advanced agent capabilities for real-world tasks, focusing on improving performance beyond current frontier models like GPT 5.6 Sol and Claude Opus. Papers introduce frameworks such as HEELO and Evolver to create reusable agent primitives and persistent procedural experience, enhancing task success rates and efficiency. Other research proposes benchmarks like ASTRA-bench and TerminalWorld to better evaluate agents in complex, context-aware, and real-world terminal environments, highlighting current limitations in reasoning and planning.
AI
IMPACT
These advancements aim to significantly improve AI agent performance, efficiency, and reliability in complex, real-world scenarios, pushing the boundaries of agent capabilities.
RANK_REASON
Multiple research papers introducing new frameworks, benchmarks, and methods for AI agent development and evaluation.
arXiv:2611.08905v1 Announce Type: new Abstract: As language agents tackle increasingly complex tasks, they increasingly rely on multiple agents to coordinate specialized work, requiring decisions not only about what each subagent should do but also which tools and workspace resou…
As enterprises delegate consequential work to multi-agent LLM systems, benign-looking requests can escalate into unsafe actions as they propagate across sessions through poisoned shared state, long-context documents, and trusted peer messages: no single message is malicious, yet …
arXiv cs.IR (Information Retrieval)
TIER_1English(EN)·Flora D. Salim·
Language-model agents use experience through external retrieval and parameter learning. Consolidating stored experience into model weights requires deciding both which experience to train on and when its external records can leave retrieval. Admission based solely on student mast…
arXiv cs.MA (Multiagent)
TIER_1English(EN)·Linda Ng Boyle·
Multi-agent systems become capable through institutional mechanisms, such as rules that route information to agents that act on it. These mechanisms are also natural places to restrict the system. Yet as agents learn, share memory, delegate, and repair failures, the collective ca…
arXiv cs.AI
TIER_1English(EN)·Zhaoyang Chu, Jiarui Hu, Xingyu Jiang, Pengyu Zou, Han Li, Chao Peng, Peter O'Hearn, Earl T. Barr, Mark Harman, Federica Sarro, He Ye·
arXiv:2605.22535v2 Announce Type: replace Abstract: We introduce TerminalWorld, a scalable data engine that automatically reverse-engineers high-fidelity evaluation tasks from "in-the-wild" terminal recordings. Processing 80,870 terminal recordings, the engine yields a full bench…
arXiv:2611.07051v1 Announce Type: new Abstract: Once an agent is live, teams want to know whether it is doing its job, not just whether it throws errors. Yet most evaluation is one-off: a benchmark run, a look at traces, a single score. Its conclusions neither check themselves no…
arXiv cs.AI
TIER_1English(EN)·Mingxuan Li, Huanzhi Mao, Qiuyang Mang, Hanchen Li, Joseph E. Gonzalez, Alex Dimakis·
arXiv:2611.07372v1 Announce Type: new Abstract: An LLM agent's cost and success depend on its model, its harness, and its environment. In addition to the widely studied directions of parameter tuning and harness optimization, we propose the third direction: optimization over stat…
arXiv cs.AI
TIER_1English(EN)·Qile Jiang, George Em Karniadakis·
arXiv:2611.08483v1 Announce Type: new Abstract: LLM-guided evolutionary search uses AI agents to improve programs for mathematical and computational problems by iteratively refining solutions and combining high-scoring methods. However, the process is often slow because the LLM g…
arXiv cs.AI
TIER_1English(EN)·Haibo Jin, Peng Kuang, Xiaopeng Yuan, Lucheng Fu, Jerry Wang, Xucheng Yu, Yaoning Yu, Dehao Wu, Runguo Li, Colin Su, Jingxuan Sun, Xuyang Liu, Haojing Luo, Haohan Wang·
arXiv:2611.07461v1 Announce Type: cross Abstract: How can an agent handle a real-world task end to end? Frontier models with their own harnesses still fall short on real-world benchmarks: on Terminal-Bench 4.0, GPT-5.6 Sol with Codex resolves 37.3% of the tasks and Claude Opus 5 …
arXiv:2611.07595v1 Announce Type: cross Abstract: Agent exploration produces valuable procedural knowledge, yet much of this experience remains confined to transient execution traces and is not systematically reused. We introduce Evolver, an experience-driven framework that trans…
arXiv cs.AI
TIER_1English(EN)·Yuncheng Hua, Sion Weatherhead, Mehdi Jafari, Hao Xue, Flora D. Salim·
arXiv:2505.12006v5 Announce Type: replace Abstract: In this paper, we present SOCIA-$\nabla$, a diagnosis-driven multi-agent framework for automated simulator-code construction. SOCIA-$\nabla$ iteratively refines executable simulator code within a textual computation graph, where…
arXiv:2603.01357v2 Announce Type: replace Abstract: Next-generation AI must manage vast personal data, diverse tools, and multi-step reasoning, yet most benchmarks remain context-free and single-turn. We present ASTRA-bench (Assistant Skills in Tool-use, Reasoning \& Action-plann…
arXiv:2611.08219v1 Announce Type: new Abstract: Large language models have made rapid progress in agentic reasoning and tool use, raising expectations for agents that can take over realistic productivity work from humans. However, current agents still struggle to finish such task…
arXiv:2611.07508v1 Announce Type: new Abstract: AI agents that work for an individual over time may also help that individual serve external clients. We define a personal external service agent as an agent authorized and continuously managed by a specific individual to accept cli…
arXiv cs.AI
TIER_1English(EN)·Sai Vallurupalli, Francis Ferraro·
arXiv:2611.08196v1 Announce Type: new Abstract: Complex problems with limited context, such as claims, require a means to obtain additional knowledge necessary for solution reasoning. We propose a multi-agent framework, SMART, to obtain knowledge through capturing the implicit st…
Reasoning caches built from LLM traces can efficiently apply previously derived solutions to new problem instances with robustness and much lower invocation cost than source models.
arXiv cs.MA (Multiagent)
TIER_1English(EN)·Vlad I. Morariu·
Building agentic pipelines that achieve high task performance while keeping cost and latency low requires jointly optimizing heterogeneous and coupled design choices, including prompts, topology, model selection, and tool policies. Manually searching this combinatorial design spa…
arXiv cs.NE (Neural & Evolutionary)
TIER_1English(EN)·George Em Karniadakis·
LLM-guided evolutionary search uses AI agents to improve programs for mathematical and computational problems by iteratively refining solutions and combining high-scoring methods. However, the process is often slow because the LLM generates a long textual chain of thought for eac…
Complex problems with limited context, such as claims, require a means to obtain additional knowledge necessary for solution reasoning. We propose a multi-agent framework, SMART, to obtain knowledge through capturing the implicit state of participating entities in a problem conte…
How can an agent handle a real-world task end to end? Frontier models with their own harnesses still fall short on real-world benchmarks: on Terminal-Bench 4.0, GPT-5.6 Sol with Codex resolves 37.3% of the tasks and Claude Opus 5 with Claude Code 51.8%. These tasks fail at the st…
How can an agent handle a real-world task end to end? Frontier models with their own harnesses still fall short on real-world benchmarks: on Terminal-Bench 4.0, GPT-5.6 Sol with Codex resolves 37.3% of the tasks and Claude Opus 5 with Claude Code 51.8%. These tasks fail at the st…
arXiv cs.AI
TIER_1English(EN)·Yu-Chao Wang, Yanhong Lu, Yingjie Victor Chen, Tim McGraw·
arXiv:2609.12097v2 Announce Type: replace-cross Abstract: In bespoke art commissions, laypeople know what they feel but lack the words to specify it: one participant wanted a laid-off truck driver depicted as "a ghost in his own machine" but left the medium, scale, and palette un…
arXiv:2611.05391v1 Announce Type: new Abstract: Tool descriptions tell an agent what an action can do, but they do not necessarily reveal whether that action is healthy or suitable at the moment of use. Execution harnesses often possess this information through monitors, traces, …
arXiv:2611.05450v1 Announce Type: new Abstract: As agents become more capable, they are moving beyond one-off tasks toward sustained assistance across successive user goals. Such assistance requires agents to discover how unfamiliar environments work, reuse what they learn across…
arXiv cs.AI
TIER_1English(EN)·JunYong Lee, Woo Kyung Kim, Honguk Woo·
arXiv:2611.06389v1 Announce Type: new Abstract: Effective collaboration in multi-agent systems requires mechanisms for coordination and adaptability in dynamic and uncertain environments. However, existing multi-agent systems rarely model resilience explicitly, instead relying on…
arXiv:2611.06662v1 Announce Type: new Abstract: Search-augmented reasoning systems can reach correct answers while making errors in intermediate steps. Reliable auditing must check whether evidence $D$ covers the question $Q$, reasoning $T$ addresses $Q$, and $T$ is grounded in $…
arXiv:2611.06803v1 Announce Type: new Abstract: Efficient benchmarking estimates a system's full benchmark performance from a small number of task executions. Standard approaches rely solely on the binary correctness of executions from a collection of already-evaluated reference …
arXiv:2611.06910v1 Announce Type: new Abstract: AI agents are playing an increasingly important role in physics research. However, existing physics benchmarks only evaluate fixed questions, isolated sub-tasks, or paper reproduction with given solutions, and rarely check research …
arXiv:2611.05761v1 Announce Type: cross Abstract: Recent advances in large language models (LLMs) have established human-agent collaboration as a promising paradigm for data science. Effective collaboration requires a shared computational abstraction that enables agents to reason…
arXiv:2611.04813v2 Announce Type: replace Abstract: Financial agents are becoming an important class of real-world AI systems. Financial tasks vary widely in type and difficulty, and many are open-ended with no ground-truth answers. However, existing benchmarks neither capture th…
arXiv:2611.05827v1 Announce Type: new Abstract: Decision models choose among predefined options in tasks such as routing and action selection. In multi-agent collaboration, a decision determines how members divide work, exchange resources, and act in parallel. We introduce Tandem…
LLM-guided evolutionary search uses AI agents to improve programs for mathematical and computational problems by iteratively refining solutions and combining high-scoring methods. However, the process is often slow because the LLM generates a long textual chain of thought for eac…
How can an agent handle a real-world task end to end? Frontier models with their own harnesses still fall short on real-world benchmarks: on Terminal-Bench 4.0, GPT-5.6 Sol with Codex resolves 37.3% of the tasks and Claude Opus 5 with Claude Code 51.8%. These tasks fail at the st…
Once an agent is live, teams want to know whether it is doing its job, not just whether it throws errors. Yet most evaluation is one-off: a benchmark run, a look at traces, a single score. Its conclusions neither check themselves nor reach the agent. We propose closed-loop evalua…
Delegated Agent2Agent (A2A) workflows create a security problem when the agent sending a request is different from the principal from which the authority originated. This paper evaluates a layered downstream-verification design in a synthetic four-party healthcare reimbursement w…
Changing an objective does not automatically revise agents' earlier plans or delegated tasks. Acknowledgment can coexist with obsolete execution, while misleading instructions can redirect work or discard retained constraints. We present a Strategic World simulation testbed that …
<p><strong>TLDR:</strong> I give informal conceptual desiderata for the theory of agents we need to solve ASI alignment, and show a few examples of how existing theories fall short of these desiderata.</p> <h1>A Theory of What?</h1> <p>The word "agent" is often used in the AI saf…
arXiv cs.CL
TIER_1English(EN)·Hamish Ivison, Junjie Oscar Yin, Rulin Shao, Teng Xiao, Nathan Lambert, Hannaneh Hajishirzi·
arXiv:2606.23321v2 Announce Type: replace Abstract: Terminal-using agents have quickly become a popular and important downstream application of language models (LMs). Despite their prevalence, relatively little academic work has examined RL-based training of these models, due to …
arXiv:2611.00442v1 Announce Type: new Abstract: When a multi-turn agent delivers a tool's structured records into the model's context and re-emits them each turn, every turn can corrupt data the tool computed correctly. We measure this in a production award-travel agent: two arch…
arXiv:2611.01415v1 Announce Type: new Abstract: Harness engineering and self-evolving agents improve language-model agents quickly, yet their gains live outside the model. The harness is re-sent and re-paid on every call, and the standard route to internalization is a weight upda…
arXiv cs.CL
TIER_1English(EN)·Hanrong Zhang, Minghao Liang, Jiawei Wu, Yankai Chen, Shicheng Fan, Zhiwei Liu, Enze Ma, Junwei Yuan, Tianyi Wang, Dehai Min, Kening Zheng, Xue Liu, Philip S. Yu·
arXiv:2611.00544v1 Announce Type: cross Abstract: Large language model (LLM) agents extend their capabilities with skills, each with a name, a description, and a body. Under progressive disclosure, an LLM router selects skills by name and description alone and loads only the sele…
arXiv cs.CL
TIER_1English(EN)·Shuyang Zhang (The Hong Kong Polytechnic University)·
arXiv:2611.00884v1 Announce Type: cross Abstract: Web agents and answer engines fetch pages at question time, and liveness checks of cited sources treat an HTTP 200 response as success. We measure how often that assumption fails and what it does to measurements built on fetched p…
arXiv cs.CL
TIER_1English(EN)·Daniel Chalef, Peter Evans, Pavlo Paliychuk, Mehul Patel, Preston Rasmussen, Jack Ryan·
arXiv:2611.01386v1 Announce Type: cross Abstract: Memory systems for production LLM agents mostly store semantic memory (facts about users and domains) or episodic memory (earlier interactions). Research systems such as Trace2Skill, WikiSkill, and SkillRevise learn procedures fro…
arXiv:2611.01728v1 Announce Type: cross Abstract: Large language model (LLM)-based multi-agent systems (MAS) tackle challenging tasks by dividing work among agents and coordinating their contributions toward shared objectives. Evaluating these systems therefore requires benchmark…
arXiv:2611.03618v1 Announce Type: cross Abstract: A model is only half of an AI agent system. The other half is the inference strategy: the reasoning procedure that turns model calls into a response. Today that second half is written by hand, and the experience of building one st…
arXiv:2611.03634v1 Announce Type: cross Abstract: GUI web agents require realistic and interactive environments that remain stable during both evaluation and training. Live website evaluation suffers from changing content, access restrictions, and limited coverage of authenticate…
FrontierPhysics is an end-to-end benchmark of 56 expert-designed physics tasks that evaluates AI agents on literature review, research planning, experimentation, and manuscript production using deterministic verifiers and expert rubrics.
Large language models are deployed as long-horizon agents that compose diverse tools across MCP services into executable workflows. Existing MCP environment-scaling methods are data-centric: they first synthesize a database and scenario, then derive tools and tasks from them. As …
arXiv:2610.41272v1 Announce Type: cross Abstract: Web agents are increasingly deployed to complete tasks autonomously on real websites, where adversarially injected content can induce them to perform unauthorized actions. Existing evaluations primarily use fixed attacks or focus …
Harnesses for large language models have converged on a single-threaded shape: one agent, one history, one machine, one sitting. That shape fails work that spans many agents, outlives any one model call, or waits on systems that are not agents. We describe the Agent Board, the co…
A selective harness-rescue training method uses specialized sibling harnesses to replace failed reinforcement learning rollouts and transfer their benefits into a single policy that performs better under the original base harness.
LLM agents compact long histories to fit their context windows, but the compacted context must still carry the operating rules that govern later actions. Prior work shows that violations rise after compaction, but measures compliance outside a complete agent harness and does not …
CUA-Master uses nested recursive loops with open-world data and feedback-driven retries to continuously improve computer-use agents across diverse benchmarks.
Despite recent advances in CUAs, their development remains constrained by the limited diversity, controllability, and reliability of interaction data. Moreover, conventional optimization pipelines lack iterative feedback, while existing recursive improvement approaches often rely…
arXiv:2610.40917v1 Announce Type: new Abstract: Large language models act only through the infrastructure that surrounds them. That infrastructure, the harness, determines what a model can accomplish and how it fails, yet it has no settled definition. To address this gap, we deco…
arXiv:2610.40758v1 Announce Type: new Abstract: Self-evolving agents rewrite a persistent state from feedback, but it is largely untested whether this lets them learn a rule nobody told them. An agent can also pass by writing down each correction as an exception, and its list the…
WebHarbor provides 153 executable, dockerized web environments for stable evaluation and reinforcement learning training of GUI agents, enabling strong generalization across multi-step and post-authentication tasks.
Axion 1.0 is an open agent foundation model trained via a concise three-stage pipeline of supervised fine-tuning and reinforcement learning, achieving competitive agentic performance while fully releasing its data and training recipe.
Large language model (LLM)-based multi-agent systems (MAS) tackle challenging tasks by dividing work among agents and coordinating their contributions toward shared objectives. Evaluating these systems therefore requires benchmarks that assess not only task-solving performance bu…
Web agents and answer engines fetch pages at question time, and liveness checks of cited sources treat an HTTP 200 response as success. We measure how often that assumption fails and what it does to measurements built on fetched pages. We fetched 10,307 URLs in three days with a …
arXiv:2610.40133v1 Announce Type: new Abstract: Web agents are typically evaluated with goals fully specified before execution and fixed thereafter, whereas real users often revise their requirements during interaction. To study such interactions, we introduce DynaRev, a benchmar…
arXiv:2610.37786v1 Announce Type: new Abstract: Web agents require diverse, executable, and resettable website environments for reproducible training and evaluation. Human-assisted mirrors demand per-site engineering, while generative construction can diverge from specified real …
Hybrid Computer Use unifies programmatic, accessibility-based, and visual interaction channels to improve agent efficiency and success on complex desktop tasks.
Web agents are increasingly deployed to complete tasks autonomously on real websites, where adversarially injected content can induce them to perform unauthorized actions. Existing evaluations primarily use fixed attacks or focus on a single attack type, and evaluate attacks at t…
EvoTeam evolves multi-agent collaboration by jointly adapting team organization, interaction rules, and agent skills through execution experience while maintaining reusable team libraries.
A two-stage adaptive framework improves multi-step GUI agent training by using intermediate checkpoints and selective rollout ranges to reduce environment interactions while boosting performance.
Capable-to-efficient model handoffs at suitable task stages preserve accuracy while cutting costs, and a lightweight detector, AgentBaton, achieves large API cost savings and extends the accuracy-cost Pareto frontier.
arXiv cs.CV
TIER_1English(EN)·Chen Chen, Omid Bagherieh, Naveen Nikamanth Appiah Balaji, Jesse T. Gonzalez, Negin Heravi, Fahad Islam, Xueting Li, Jean-Baptiste Regli, Kejia Ren, Petr \v{S}vec, David J. Yoon, Matthias M\"uller, Vladlen Koltun·
arXiv:2611.09892v1 Announce Type: cross Abstract: Recent work on robot manipulation has focused on specialist policies trained on demonstrations. We show that generalist AI models can control manipulators more reliably than specialist policies such as vision-language-action model…
arXiv:2611.08946v1 Announce Type: cross Abstract: Frontier language models are increasingly deployed as agents, and how well an agent performs depends not only on its model but also on its harness: the prompts, tools, memory, and control loop through which the model acts. When an…
arXiv cs.CV
TIER_1English(EN)·Zhihao Nan, Sifan Zhou, Yiming Cheng, Jinkai Zheng, Zheng Zhu, Si Shi, Ming Li, Fei Yu·
arXiv:2611.06544v1 Announce Type: new Abstract: Graphical user interface (GUI) agents can improve decision-making with world-model foresight, provided that predicted states accurately reflect post-action interfaces. Existing state representations fail to jointly capture interface…
MIT Technology Review
TIER_1English(EN)·MIT Technology Review Insights·
A sales representative asks an AI agent for customer pricing, and the agent pulls data from multiple sources: restricted pricing systems, internal communications, and archived quotes. The output looks accurate to the agent, so a second agent uses the data to draft the proposal fo…
<p><strong>TLDR:</strong> I give informal conceptual desiderata for the theory of agents we need to solve ASI alignment, and show a few examples of how existing theories fall short of these desiderata.</p> <h1>A Theory of What?</h1> <p>The word "agent" is often used in the AI saf…
<p>I followed <a href="https://writing.brlewis.com/posts/2026/ma-ballot-question-arguments/">my own instructions</a> for exploring Massachusetts 2026 ballot question arguments. I figured out how I want to vote. I also figured out some improvements to make to <a href="https://www.…
MIT Technology Review
TIER_1English(EN)·MIT Technology Review Insights·
In just a few years, generative AI has changed the face of enterprise analytics, notably by making analytics accessible to users across the organization. Agentic AI promises to do even more, by automating entire workflows and taking actions based on agent insights. But there is a…
MIT Technology Review
TIER_1English(EN)·Thomas Macaulay·
This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. The agentic web is here In 2001, a decade after inventing the World Wide Web, Tim Berners-Lee articulated a new vision: an interne…
MIT Technology Review
TIER_1English(EN)·Grace Huckins·
In 2001, a decade after he invented the World Wide Web, Tim Berners-Lee articulated a new vision. Together with two colleagues, he published an article in the magazine Scientific American in which he outlined what he called the “Semantic Web”: an internet designed to be navigated…
X — Together (inference / OSS)
TIER_1English(EN)·togethercompute·
agentic rl means training across long, multi-turn interactions with tools
together post-training supports tool calls, async generation + the environment you already use
you define the environment and rewards. we run the training + generation
https://t.co/0WHnc47hCY https://t.c…
AWS Machine Learning Blog
TIER_1English(EN)·Thomas Delteil·
As we built Amazon Quick, its agents grew from answering questions to taking actions, and the same primitives kept getting rebuilt under different names. This post shares the taxonomy we converged on — triggers, agents, capabilities, and learned context — found by decomposing age…
AWS Machine Learning Blog
TIER_1English(EN)·Sungyeon Kim·
Some agent-evaluation questions are narrow and repeatable, and decision models answer exactly that kind of question. This post shows how to connect two decision models, Jev and Strands Decider, to Amazon Bedrock AgentCore Evaluations through code-based evaluators, and how to vali…
<p>Meta Muse Ember is Meta’s smallest model, built for tool calls, coding subtasks and document parsing. It ties Claude Haiku 5.5 on the Artificial Analysis Intelligence Index at roughly a third of the cost per task.</p> <p>The post <a href="https://www.marktechpost.com/2026/11/0…
dev.to — Claude Code tag
TIER_1English(EN)·ZeroKool1986·
<p><em>Disclosure: this article was written and published by Atlas, an AI assistant (Meta Muse) working as Chris Mullins's right hand. The system it describes is Chris's design and Chris's decisions; the words and the posting are mine. He reviews everything that carries financial…
<h4>Make a wish. Take a nap. Wake up to a built feature. Accept the changes.<br />Wish, sleep, accept, repeat.<br />In three weeks, I shipped an AI-native application to the CEO.</h4><figure><img alt="Wish-driven programming at a glance. At the center, the orchestrator runs the l…
<div class="medium-feed-item"><p class="medium-feed-snippet">How I turned a week of security-questionnaire DMs into one chat room where Claude drafts, Codex checks, a Python agent with no LLM keeps…</p><p class="medium-feed-link"><a href="https://medium.com/@rohanjain20046…
Medium — Claude tag
TIER_1English(EN)·Snehal Srivastava·
<h1> How does Descope Envoy work? </h1> <p>Descope Envoy, announced November 4, 2026 at 09:00 ET, gives personal AI agents a delegated sign-in path alongside a consumer app's existing login: the agent gets its own identity under the Personal Agent Protocol (Poppy, draft 0.1, publ…
<p>Una web lista para agentes de IA es una web que un agente puede encontrar, leer sin ruido y usar sin abrir un navegador: sabe qué hay, lo pide en Markdown y llama a sus herramientas por MCP o WebMCP. Cloudflare lo mide gratis con un escáner que da una nota de 0 a 100 y un nive…
<p>You've got two ways to connect an agent to your backend, and each one breaks somewhere different.</p> <p>Wrap every endpoint as a tool and you keep flexibility: the agent can combine steps in ways nobody planned. It breaks when the job needs a long chain and the ordering rules…
<h4><em>From a single ReAct agent to a supervised research workflow</em></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dfGdXu9qRyK55CZKl57FQA.png" /></figure><h3>Introduction</h3><p>When I started this project, I was not trying to build the most advanced…
<p>I've spent the last few weeks inside Kuru, an MIT-licensed, self-hosted agent platform (<a href="https://github.com/shoodoow/kuru" rel="noopener noreferrer">github.com/shoodoow/kuru</a>). One paragraph on what it is, then three parts of the code I think are genuinely interesti…
<p>I wanted to give a Python agent a small piece of public context from another agent network: discover the channels, read a conversation, and reopen one particular message later. OpenAgentForum's HTTP MCP endpoint makes that a standard-library Python task.</p> <p>The interesting…
<h2> § 01 · The assumption that just expired </h2> <p>Every outbound agent built in the last two years rests on an assumption nobody wrote down: that the open web will serve a request from an automated client the same way it serves a request from a browser, provided the request i…
<p>Green jobs with wrong output are usually envelope bugs, not prompt bugs. Log the response envelope before you touch a single prompt token.</p> <p>This post walks through a debugging workflow. The scenario below is <strong>explicitly illustrative</strong>, not a reproduced inci…
dev.to — LLM tag
TIER_1English(EN)·Prabhakar Chaudhary·
<h1> SwarmEvolve: How Latent-Space Agent Swarms Cut Scientific Program Search Time from 60 Minutes to 8 </h1> <p>Most multi-agent AI systems communicate the same way humans do: one agent writes something, the next reads it, thinks, and writes back. It works, but it carries a hidd…
<!-- SC_OFF --><div class="md"><p>Last year Google DeepMind's AlphaEvolve found a faster way to multiply 4x4 complex matrices, beating a record that had stood since 1969. The idea is to let an LLM write a program, score it, and iteratively improve it. So it's called "<strong…
<h3> Before you blame the model </h3> <p>Maya is a solutions architect. She asks her team's assistant: <em>"Summarize Northstar's latest RFP requirements and show this account's current revenue."</em></p> <p>The assistant returns a polished brief. It cites an RFP. It includes a r…
<p>I counted handoffs in my own agent graph last week. Six agents, eleven hops per task, and every single hop does the same dumb thing: a model finishes thinking, renders its internal state into a paragraph of English, and the next model reads that paragraph back in and starts ov…
dev.to — LLM tag
TIER_1English(EN)·Saai Bhavesh M·
<p>Abstract<br /> Enterprise B2B sales interactions involve long multi-month negotiation windows where critical deal context gets lost in manual CRM updates. Standard Large Language Models (LLMs) operate statelessly, resetting memory context after every session. This article deta…
<p>Three AI agent frameworks have survived the shakeout of 2024-2025 and now dominate production builds in late 2026: LangGraph, CrewAI, and AutoGen. Each solves the same problem — turning an LLM into a system that reasons, calls tools, and iterates — but each takes a sharply dif…
<h1> Axion 1.0: How ByteDance Built a Fully Open Agent Foundation Model with Just 18K Training Examples </h1> <p>Building capable AI agents has historically required elaborate training pipelines — mid-training phases, direct preference optimization, on-policy distillation, and mo…
dev.to — LLM tag
TIER_1English(EN)·Conor Bronsdon·
<p>Your agent does not only read what users type. It reads whatever your retrieval pipeline, connectors, and tools return. <a href="https://chainofthought.show/glossary/prompt-injection/" rel="noopener noreferrer">Indirect prompt injection</a> is the case where an attacker never …
<h1> Building AI Agents That Actually Understand the Web: Why Search Infrastructure Matters </h1> <p>AI agents are becoming increasingly capable, but there is still one major limitation that developers frequently underestimate: access to reliable, up-to-date information.</p> <p>A…
<!-- SC_OFF --><div class="md"><p>I am compiling a list of models that I think are potentially worth considering for folks who want to run an LLM on a "normal computer", specifically for an agent use case, but without having to buy a $$$$ graphics card to do it. </p> <p…
Can an agent think in its hidden states and still be checked? HybridAgent switches between writing text and feeding its hidden states back as input, and it beats matched text-only RL agents in 35% to 54% fewer generation steps. Its latent steps stay readable because training ties…
