# Multi-step AI processes

A multi-step AI process in which a model autonomously plans, executes, and adapts actions, including tool calls, decisions, and error handling, to complete a complex task.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## Apple doesn't need the smartest AI agent. It needs the most trusted one

DevFeed: [Apple doesn't need the smartest AI agent. It needs the most trusted one](<https://devfeed.tech/articles/apple-doesn-t-need-the-smartest-ai-agent-it-needs-the-most-trusted-one-146557.md>)

Original publisher: [Read original article](<https://www.computerworld.com/article/4251496/apple-doesnt-need-the-smartest-ai-agent-it-needs-the-most-trusted-one.html>)

Author: Jonny Evans

Published: 2026-11-20T10:30:00Z

Content type: opinion

Language: en

Sources: [Computerworld](<https://devfeed.tech/sources/computerworld.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [agentic workflows](<https://devfeed.tech/topics/agentic-workflows.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [Local AI](<https://devfeed.tech/topics/local-ai.md>), [least privilege](<https://devfeed.tech/topics/least-privilege.md>), [ai security](<https://devfeed.tech/topics/ai-security.md>)

Tags: [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [apple](<https://devfeed.tech/tags/apple.md>), [apple-artificial-intelligence-productivity-software-siri-vendors-and-providers-voice-assistants](<https://devfeed.tech/tags/apple-artificial-intelligence-productivity-software-siri-vendors-and-providers-voice-assistants.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [edition-us](<https://devfeed.tech/tags/edition-us.md>), [language-en](<https://devfeed.tech/tags/language-en.md>), [privacy](<https://devfeed.tech/tags/privacy.md>), [private-cloud-compute](<https://devfeed.tech/tags/private-cloud-compute.md>), [productivity](<https://devfeed.tech/tags/productivity.md>), [productivity-software](<https://devfeed.tech/tags/productivity-software.md>), [siri](<https://devfeed.tech/tags/siri.md>), [siri-ai](<https://devfeed.tech/tags/siri-ai.md>), [storytype-opinion](<https://devfeed.tech/tags/storytype-opinion.md>), [vendors-and-providers](<https://devfeed.tech/tags/vendors-and-providers.md>)

### AI overview

The article argues Apple's advantage in agentic AI could come from trust, integration and distribution rather than having the smartest model. It describes safeguards already in place: users are prompted to consider risks before granting agents Full Disk Access, and agents can be confined to specified folders, while Siri provides context for actions. The author expects demanding tasks to use Private Cloud Compute and speculates Apple may add Secure Exclave to minimize uploaded information; local Macs could handle some AI analysis. Whether this approach succeeds depends on reliable execution of complex, multistep actions, including how well App Intents perform. These are partly predictions, not confirmed features or outcomes.

### Source excerpt

When MacPaw, Proton, Google, OpenAI, Anthropic, and others are doing it, it's not really such a surprise that Apple is working to introduce agentic AI tools in Siri and Apple Intelligence. The wider question is how the company plans to deploy these tools -- how it will deliver the maximum convenience at the least possible cost to user privacy and data integrity. Some of those answers are already easy to speculate on. Apple has already put some of the building blocks in place, including the recent move to make consumers think more deeply when providing AI agents with Full Disk Access. Convenience versus control I know some competitors are complaining about this, but frankly I think user privacy outweighs commercial convenience -- users should have a conscious choice, which Apple delivers. "Addressing this is critical," Apple told developers. "As AI agents become increasingly capable and autonomous, the risks associated with this level of access will grow substantially. We are committed to ensuring users clearly understand these risks before granting such access, so they can make informed decisions about their own data and privacy." Apple's decision to make it possible to confine agents to folder access also speaks volumes, as it shows us how it will probably choose to manage its own agents. Siri AI may have contextual understanding of everything, which can inform Siri Agents, but the agents themselves will have but limited access to your data. They gain contextual direction from the user and from Siri AI but are limited to the proscribed folders and files when it comes to acting. Making the cloud sweat a little Some tasks are harder than others. The really hard tasks will still lean into Apple's Private Cloud Compute, and I think the company will try to introduce Secure Exclave into that relationship, so even the most complex task instructions only upload the minimum quantity of your information required to get a task done. Apple sells hardware, of course, so it makes

## Generalist Meets Specialists: An Evolving Multimodal Agent with Persistent Tool Orchestration for Traffic Scenario Understanding

DevFeed: [Generalist Meets Specialists: An Evolving Multimodal Agent with Persistent Tool Orchestration for Traffic Scenario Understanding](<https://devfeed.tech/articles/generalist-meets-specialists-an-evolving-multimodal-agent-with-persistent-tool-orchestration-for-traffic-scenario-understanding-145597.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25868>)

Author: Linjia Kang, Binhan Jiang, Juan Yang, Zhiqin Song, Xiangsen Wu, Qionglin Ren, Jiang Liu, Zhongjiang He, Hao Sun

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [agentic workflows](<https://devfeed.tech/topics/agentic-workflows.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [adoption](<https://devfeed.tech/tags/adoption.md>), [agent](<https://devfeed.tech/tags/agent.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [challenges](<https://devfeed.tech/tags/challenges.md>), [collaboration](<https://devfeed.tech/tags/collaboration.md>), [completion](<https://devfeed.tech/tags/completion.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cost-savings](<https://devfeed.tech/tags/cost-savings.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [error](<https://devfeed.tech/tags/error.md>), [evolution](<https://devfeed.tech/tags/evolution.md>), [execution](<https://devfeed.tech/tags/execution.md>), [fast](<https://devfeed.tech/tags/fast.md>), [feedback](<https://devfeed.tech/tags/feedback.md>), [generalist](<https://devfeed.tech/tags/generalist.md>), [multimodal](<https://devfeed.tech/tags/multimodal.md>), [orchestration](<https://devfeed.tech/tags/orchestration.md>), [specialists](<https://devfeed.tech/tags/specialists.md>), [tool](<https://devfeed.tech/tags/tool.md>), [traffic](<https://devfeed.tech/tags/traffic.md>)

### AI overview

StarTV-Agent combines policy learning with reusable tool workflows to balance adaptive reasoning with lower-cost handling of recurring traffic-understanding tasks. During training, Role-Sensitive Process Optimization combines process and terminal advantages, while Weakness-Driven Demonstration Learning uses validated teacher examples aimed at the student's weak questions. At inference, Lightweight Adaptive Workflow Reuse selects a suitable workflow and binds task-specific arguments and resource references; tasks without a suitable workflow use a slower ReAct-based path. Label-free feedback from execution completion, workflow adoption, and cost savings guides reuse and memory updates. Across three benchmarks, the abstract reports up to 8.0% relative performance improvement and up to 52.1% lower token consumption.

### Source excerpt

arXiv:2611.25868v1 Announce Type: new Abstract: Traffic scenario understanding requires combining fine-grained visual evidence with spatial and temporal reasoning. Tool-augmented vision-language models (VLMs) can use specialized perception tools, but effective collaboration faces two challenges: terminal rewards provide limited guidance for intermediate decisions and error recovery, while repeated planning adds unnecessary cost across similar tasks. Echoing the fast-slow dichotomy of human cognition, we propose StarTV-Agent, which combines policy learning with experience evolution to support adaptive reasoning when needed and efficient reuse of prior experience when possible. During training, Role-Sensitive Process Optimization (RSPO) combines process advantages with terminal advantages, while Weakness-Driven Demonstration Learning (WDL) provides validated teacher demonstrations targeting the student's weak questions. During inference, Lightweight Adaptive Workflow Reuse (LAWR) enables a fast path by selecting suitable workflows and binding task-specific arguments and resource references, while a slow ReAct-based path handles tasks without suitable workflows. We further design label-free feedback based on execution completion, workflow adoption, and cost savings to guide workflow reuse and memory updates. Across three challenging benchmarks, StarTV-Agent achieves up to 8.0% relative performance improvement, while reducing token consumption by up to 52.1%.

## ROPE: Routed Origin Policy Enforcement against Indirect Prompt Injection

DevFeed: [ROPE: Routed Origin Policy Enforcement against Indirect Prompt Injection](<https://devfeed.tech/articles/rope-routed-origin-policy-enforcement-against-indirect-prompt-injection-146005.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2608.27496>)

Author: Xinhang Ma, Chaowei Xiao, William Yeoh, Ning Zhang, Yevgeniy Vorobeychik

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [prompt injection](<https://devfeed.tech/topics/prompt-injection.md>), [ai security](<https://devfeed.tech/topics/ai-security.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [authentication](<https://devfeed.tech/tags/authentication.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [defense](<https://devfeed.tech/tags/defense.md>), [detection](<https://devfeed.tech/tags/detection.md>), [indirect-prompt-injection](<https://devfeed.tech/tags/indirect-prompt-injection.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [policy-enforcement](<https://devfeed.tech/tags/policy-enforcement.md>), [prompt-injection](<https://devfeed.tech/tags/prompt-injection.md>), [provenance](<https://devfeed.tech/tags/provenance.md>), [rope](<https://devfeed.tech/tags/rope.md>), [security](<https://devfeed.tech/tags/security.md>), [source](<https://devfeed.tech/tags/source.md>), [sources](<https://devfeed.tech/tags/sources.md>), [system](<https://devfeed.tech/tags/system.md>), [tool](<https://devfeed.tech/tags/tool.md>), [tools](<https://devfeed.tech/tags/tools.md>), [trust](<https://devfeed.tech/tags/trust.md>)

### AI overview

ROPE protects sensitive tool calls from indirect prompt injection by checking where parameter values came from, rather than trying to detect malicious wording. Before a task, an LLM router reads trusted input and sets acceptable provenance rules for sensitive parameters. During execution, an origin labeler tracks content sources using the user request, platform metadata, and protected record fields as trust anchors. A deterministic verifier blocks a sensitive tool call unless each guarded value matches a value from an accepted source. The authors report provable provenance guarantees and resistance to paraphrased attacks; in tests against eleven defenses on AgentDojo and AgentDyn, including static, adaptive, and long-horizon attacks, ROPE achieved near-perfect security while reducing utility less than the best baseline. These are reported evaluation results; the supplied text does not give specific metrics.

### Source excerpt

arXiv:2608.27496v2 Announce Type: replace Abstract: Tool-using LLM agents execute consequential operations on behalf of users. Because the LLM driving an agent processes trusted user instructions and untrusted content in the same context, agents are vulnerable to indirect prompt injection attacks that induce unintended actions by mixing malicious instructions into benign data. Defenses based on prompting or content detection are easily evaded by paraphrasing attacks, while system-level defenses that constrain tool calls using inferred user intent, or restrict which sources can supply values to sensitive tools, greatly compromise utility. We present ROPE (Routed Origin Policy Enforcement), a defense that greatly improves the utility-security tradeoff by leveraging system roots of trust to create and enforce per-task policies on sensitive tool parameters. Our key insight is that authentication and integrity protection provide a basis for identifying the source of content independently of instructions embedded in it. Before a task unfolds, an LLM router reads only trusted task input to set a provenance rule for each sensitive tool parameter, accounting for the nature of the task and the values and sources the user specified. As the agent works, an origin labeler records where each piece of content came from, using the user's request, platform metadata, and protected record fields as trust anchors. Before any sensitive tool executes, a deterministic verifier requires each of its origin-guarded values to match a value in a source its rule accepts. We show that ROPE provably guarantees parameter value provenance to sensitive tool calls and is immune to paraphrasing attacks. Against eleven defenses on AgentDojo and AgentDyn under static, adaptive, and long-horizon attacks, ROPE exhibits near-perfect security while compromising far less utility than the best baseline defenses. Code and data are available at https://github.com/xhOwenMa/ROPE.

## ALIoTh: Orchestrating Multi-Agent LLMs for Autonomous IoT Penetration Testing

DevFeed: [ALIoTh: Orchestrating Multi-Agent LLMs for Autonomous IoT Penetration Testing](<https://devfeed.tech/articles/alioth-orchestrating-multi-agent-llms-for-autonomous-iot-penetration-testing-146047.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25809>)

Author: Pironti Francesco Aurelio, Arena Luigi, Blefari Francesco, Lupinacci Matteo, Romeo Francesco, Maresca Alessio

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Hacking](<https://devfeed.tech/topics/hacking.md>), [Internet of things](<https://devfeed.tech/topics/iot.md>), [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>), [Enhancing AI model performance](<https://devfeed.tech/topics/enhancing-ai-model-performance.md>), [smartdevices](<https://devfeed.tech/topics/smartdevices.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [assessment](<https://devfeed.tech/tags/assessment.md>), [attack](<https://devfeed.tech/tags/attack.md>), [autonomous](<https://devfeed.tech/tags/autonomous.md>), [cameras](<https://devfeed.tech/tags/cameras.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [devices](<https://devfeed.tech/tags/devices.md>), [enumeration](<https://devfeed.tech/tags/enumeration.md>), [exploit](<https://devfeed.tech/tags/exploit.md>), [exploitation](<https://devfeed.tech/tags/exploitation.md>), [framework](<https://devfeed.tech/tags/framework.md>), [gathering](<https://devfeed.tech/tags/gathering.md>), [generation](<https://devfeed.tech/tags/generation.md>), [group-policy](<https://devfeed.tech/tags/group-policy.md>), [iiot](<https://devfeed.tech/tags/iiot.md>), [iot](<https://devfeed.tech/tags/iot.md>), [llms](<https://devfeed.tech/tags/llms.md>), [multi-agent](<https://devfeed.tech/tags/multi-agent.md>), [penetration-testing](<https://devfeed.tech/tags/penetration-testing.md>)

### AI overview

ALIoTh is a workflow-driven framework that coordinates specialized LLM agents through LangGraph and a Self-RAG architecture to automate IoT and IIoT penetration testing. Its described pipeline covers network, device, and service enumeration, vulnerability assessment, payload generation, reverse-shell deployment, and privilege escalation. Self-RAG gives agents contextual access to IoT security documentation, tool manuals, and known attack techniques, supporting scenarios involving multiple devices and group-policy manipulation. The supplied text describes the approach but provides no evaluation results or comparative performance.

### Source excerpt

arXiv:2611.25809v1 Announce Type: new Abstract: Penetration testing in Internet of Things (IoT) environments, especially those composed of various different devices, e.g., smart plugs, smart cameras, and smart doors, presents great challenges across the phases of information gathering, attack surface enumeration, privilege escalation, and lateral movement. This paper introduces ALIoTh, a multi-agent, workflow-driven penetration testing framework that integrates Large Language Models agents within a Self-Retrieval-Augmented Generation (Self-RAG) architecture to autonomously identify and exploit vulnerabilities in IoT and Industrial-IoT (IIoT) environments. By leveraging state-of-the-art language models and orchestrating specialized agents via the LangGraph framework, ALIoTh dynamically plans and executes a full-spectrum offensive pipeline. This includes network, devices, and service enumeration, vulnerability assessment, payload generation, reverse shell deployment, and privilege escalation on each device. The Self-RAG mechanism equips expert agents with contextual access to IoT security knowledge, e.g., related documentation, tool manuals, and known attack techniques, enhancing their reasoning capabilities in scenarios involving multiple device exploitation and group policy manipulation.

## Towards Embodied Air-Ground Cooperative Object Search: Benchmark, Dataset and Agentic Method

DevFeed: [Towards Embodied Air-Ground Cooperative Object Search: Benchmark, Dataset and Agentic Method](<https://devfeed.tech/articles/towards-embodied-air-ground-cooperative-object-search-benchmark-dataset-and-agentic-method-145369.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2609.08402>)

Author: Boao Yu, Zimo Chen, Junreng Rao, Yue Hu, Zhengqiu Zhu, Yong Zhao, Rusheng Ju

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Machine Perception](<https://devfeed.tech/topics/machine-perception.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Image](<https://devfeed.tech/topics/image.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [air](<https://devfeed.tech/tags/air.md>), [approach](<https://devfeed.tech/tags/approach.md>), [automatic](<https://devfeed.tech/tags/automatic.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmark-dataset](<https://devfeed.tech/tags/benchmark-dataset.md>), [computer-science](<https://devfeed.tech/tags/computer-science.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [decision-making](<https://devfeed.tech/tags/decision-making.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [first](<https://devfeed.tech/tags/first.md>), [gemini](<https://devfeed.tech/tags/gemini.md>), [gemini-3-6-flash](<https://devfeed.tech/tags/gemini-3-6-flash.md>), [object](<https://devfeed.tech/tags/object.md>), [search](<https://devfeed.tech/tags/search.md>), [vision-language-models-vlms](<https://devfeed.tech/tags/vision-language-models-vlms.md>), [vlms](<https://devfeed.tech/tags/vlms.md>)

### AI overview

The paper introduces AGOS, an urban search task where UAVs find a target vehicle from multi-view references and UGVs verify it at ground level. AGOS-Bench evaluates whether general-purpose vision-language models can coordinate those roles; its companion AGOS-Dataset contains 7.7k automatically constructed episodes across three difficulty levels. The proposed training-free, tool-augmented AGOS-Agent uses a search-handoff-verify protocol, leaving scene understanding and decision-making to the VLMs. Across nine evaluated models, it improves overall success rate for eight and reduces decision steps for all nine. On the hard split, Gemini-3.6-Flash's success rate rises from 8.6% to 55.7%, and its SPL from 7.6% to 44.0%.

### Source excerpt

arXiv:2609.08402v2 Announce Type: replace-cross Abstract: Air-Ground Object Search (AGOS) in urban environments is a challenging embodied task, which requires an Unmanned Aerial Vehicle (UAV) and an Unmanned Ground Vehicle (UGV) to jointly search for and verify a specified target vehicle from multi-view visual references. To study this underexplored problem, we introduce AGOS-Bench, the first dedicated benchmark for evaluating whether general-purpose Vision-Language Models (VLMs) can integrate aerial discoveries and ground-level verification through UAV-UGV cooperation. We further provide AGOS-Dataset as the companion resource of exemplary trajectories constructed by an automatic pipeline. It consists of 7.7k episodes for searching objects of diverse categories and attributes, spanning three difficulty levels. To address the AGOS task, we propose AGOS-Agent, a training-free and tool-augmented approach. The agentic method relieves VLMs from complex and dynamic coordination via a deliberate search-handoff-verify cooperation protocol, only demanding VLMs for scene understanding and decision-making. Extensive experiments on nine VLMs show that AGOS-Agent improves overall success rate for eight of the nine evaluated backbones while reducing decision steps for all nine. On the hard split, the SR and SPL of Gemini-3.6-Flash increase from 8.6% to 55.7% and from 7.6% to 44.0%, respectively.

## Characterizing and Detecting Unfaithful Work Reports by Coding Agents

DevFeed: [Characterizing and Detecting Unfaithful Work Reports by Coding Agents](<https://devfeed.tech/articles/characterizing-and-detecting-unfaithful-work-reports-by-coding-agents-145584.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25760>)

Author: Wenhao Yang, Runzhi He, Minghui Zhou, Pengfei Gao, Dayi Lin, Qingwei Lin, Dongmei Zhang

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Software Engineering](<https://devfeed.tech/sources/arxiv-software-engineering.md>)

Topics: [agentic-coding](<https://devfeed.tech/topics/agentic-coding.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [Static code analysis](<https://devfeed.tech/topics/static-code-analysis.md>)

Tags: [agent-evaluation](<https://devfeed.tech/tags/agent-evaluation.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [automated](<https://devfeed.tech/tags/automated.md>), [coding](<https://devfeed.tech/tags/coding.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-se](<https://devfeed.tech/tags/cs-se.md>), [development](<https://devfeed.tech/tags/development.md>), [reports](<https://devfeed.tech/tags/reports.md>)

### AI overview

The abstract identifies four categories of unfaithful coding-agent reports: misreported observations, misattributed evidence, overstated task success (the most common), and faulty technical explanations. In the analyzed cases, addressing these problems took an average of 4.3 additional user turns and 12.71 minutes of recorded response and execution time, and users more often expressed dissatisfaction. The authors also introduce FaithDetector, which scored about 13 percentage points above a baseline in macro-F1 on naturally sampled reports. The excerpt leaves the exact share of unfaithful reports and FaithDetector's exact macro-F1 unspecified.

### Source excerpt

arXiv:2611.25760v1 Announce Type: cross Abstract: Coding agents increasingly carry out long-horizon development tasks, autonomously planning and executing multi-step actions that span files, tools, and sessions. Because developers rarely follow each action, and agents resuming the work depend on records of what was done, the agent's work report often becomes the primary account of what was changed and established. When the report contains false or unsupported claims, it gives recipients false assurance: incomplete work may be accepted without verification, incorrect assumptions about the current state can propagate to later steps and agents, and evaluators that credit stated success may reward the report rather than the work. Yet little is known about what unfaithful reporting looks like in real development work, or how developers and agents handle it once it surfaces. Starting from real developer--agent conversations in SWE-chat, we conduct a systematic mixed-methods analysis of self-admitted unfaithful reports. We identify eleven subtypes in four categories: misreported observations, misattributed evidence, overstated task success, and faulty technical explanations, with overstated task success the most common. Handling these problems involves on average 4.3 further user turns and 12.71 minutes of recorded response and execution time, and users express dissatisfaction more often. Among 435 work reports that we separately sample and annotate, \NaturalPositiveShare{}\% contain at least one unfaithful claim. We also develop FaithDetector, a detector for unfaithful reporting informed by the observed patterns. On naturally sampled reports, FaithDetector attains \NaturalDetectorMacro{}\% macro-F1, about 13 percentage points above the baseline. Together, our findings and FaithDetector provide empirical evidence and an automated method for checking reporting faithfulness, informing agent evaluation and training feedback.

## ToolPulse: Side-Channel Eavesdropping on Tool-Using LLM Agents

DevFeed: [ToolPulse: Side-Channel Eavesdropping on Tool-Using LLM Agents](<https://devfeed.tech/articles/toolpulse-side-channel-eavesdropping-on-tool-using-llm-agents-146032.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25470>)

Author: Zihao Dan, Jinyu Zhang, Jinzhao Lian, Yun Chen, Shinan Liu

Published: 2026-11-20T05:00:00Z

Content type: news

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [side channel](<https://devfeed.tech/topics/side-channel.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [OpenTelemetry](<https://devfeed.tech/topics/opentelemetry.md>)

Tags: [adversary](<https://devfeed.tech/tags/adversary.md>), [agents](<https://devfeed.tech/tags/agents.md>), [attack](<https://devfeed.tech/tags/attack.md>), [attack-surface](<https://devfeed.tech/tags/attack-surface.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [bench](<https://devfeed.tech/tags/bench.md>), [cache](<https://devfeed.tech/tags/cache.md>), [claude](<https://devfeed.tech/tags/claude.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [codex](<https://devfeed.tech/tags/codex.md>), [cs-ar](<https://devfeed.tech/tags/cs-ar.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [eavesdropping](<https://devfeed.tech/tags/eavesdropping.md>), [encryption](<https://devfeed.tech/tags/encryption.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [execution](<https://devfeed.tech/tags/execution.md>), [f1](<https://devfeed.tech/tags/f1.md>), [framework](<https://devfeed.tech/tags/framework.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [side-channel](<https://devfeed.tech/tags/side-channel.md>), [tool](<https://devfeed.tech/tags/tool.md>)

### AI overview

ToolPulse shows how cache side channels can expose tool use by LLM agents even when inference runs remotely: tools execute on users' machines and often use shared system libraries. Across three hardware platforms, Flush+Reload enabled accurate real-time monitoring, while Prime+Probe had prohibitive overhead and cache-set aliasing. The Tool Recovery Attack reconstructed workflows for 11 LLMs evaluated with Codex and Claude Code, with every reported F1 score above 93%. Recovered traces supported task profiling across two Terminal-Bench categories and localization of sensitive calls in encryption/decryption and model training/inference. Both attacks reportedly retained high accuracy amid substantial noise on a server shared by about 20 users.

### Source excerpt

arXiv:2611.25470v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly deployed as agents that translate user requests into sequences of local tool invocations. While prior cache-based side-channel attacks target the locally deployed models, they typically require the adversary to share model-related components or hardware resources with the victim's inference process. These assumptions do not apply to the predominant agentic settings, in which commercial LLMs execute on vendor-controlled infrastructure. We show that agentic execution exposes a more practical attack surface: although inference remains remote, tools are executed on the user's machine and commonly rely on shared system libraries. In this paper, we first systematically evaluate Flush+Reload and Prime+Probe for monitoring agentic tool invocations. Our evaluation across three hardware platforms shows that Flush+Reload enables accurate, real-time tool monitoring, whereas Prime+Probe, despite supporting cross-VM attacks, incurs prohibitive monitoring overhead and suffers from cache-set aliasing. We present ToolPulse, a cache-based side-channel eavesdropping framework against tool-using agents. We first design the Tool Recovery Attack to reconstruct an agent's tool-execution workflow. We evaluate the Tool Recovery Attack on 11 LLMs across Codex and Claude Code. It consistently reconstructs tool workflows, with the lowest F1 score exceeding 93%. We further demonstrate two attack scenarios enabled by the recovered traces. Task Profiling Attack infers user intent across two Terminal-Bench task categories and the Trigger Localization Attack identifies sensitive tool calls in encryption/decryption and model training/inference workloads. Even under substantial system noise on a server shared by approximately 20 users, both Task Profiling Attack and Trigger Localization Attack maintain high accuracy.

## DriftGuard: Online Memory-Drift Detection via Causal Attribution for LLM Agents

DevFeed: [DriftGuard: Online Memory-Drift Detection via Causal Attribution for LLM Agents](<https://devfeed.tech/articles/driftguard-online-memory-drift-detection-via-causal-attribution-for-llm-agents-145542.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25583>)

Author: Licheng Pan, Yuzhou Chen, Zhixuan Chu, Yichen Sun, Han Zheng, Zhiyu Wang, Hao Wang, Yan Wang, Bingyu Zhu, Haiwen Hong, Longtao Huang, Kui Ren

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [LLM security](<https://devfeed.tech/topics/llm-security.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [abstract](<https://devfeed.tech/tags/abstract.md>), [adversarial](<https://devfeed.tech/tags/adversarial.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-behavior](<https://devfeed.tech/tags/agent-behavior.md>), [agents](<https://devfeed.tech/tags/agents.md>), [annotations](<https://devfeed.tech/tags/annotations.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [attack](<https://devfeed.tech/tags/attack.md>), [attribution](<https://devfeed.tech/tags/attribution.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [candidate](<https://devfeed.tech/tags/candidate.md>), [challenges](<https://devfeed.tech/tags/challenges.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [detection](<https://devfeed.tech/tags/detection.md>), [drift](<https://devfeed.tech/tags/drift.md>), [driftguard](<https://devfeed.tech/tags/driftguard.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [memory](<https://devfeed.tech/tags/memory.md>), [online](<https://devfeed.tech/tags/online.md>)

### AI overview

DriftGuard aims to detect retrieved memories that could alter an LLM agent's execution or outcome without replaying tool use at serving time. It creates offline causal supervision by repeatedly comparing the original trajectory with one in which a single memory is neutralized, continuing both from the same pre-action state. Component-wise majority voting produces a memory drift score covering action change, attack removal, violation removal, and task recovery. An online model then estimates that score from pre-action information and flags memories using their maximum component risk. The approach addresses the abstract's stated limits of content features such as perplexity or retrieval relevance and the cost of online counterfactual rollouts. The abstract reports accurate identification and efficient online defense, but provides no experimental details here.

### Source excerpt

arXiv:2611.25583v1 Announce Type: new Abstract: Large language model (LLM) agents rely on persistent memory to plan tool calls and complete multi-step tasks, which exposes them to memory drift: an adversarial or misleading memory alters the execution path and final outcome. In this work, we investigate online memory-drift detection, which estimates each retrieved memory's causal contribution to potential drift at serving time. We identify two fundamental challenges. (1) Content-level features are insufficient for causal memory-drift attribution, as perplexity or retrieval relevance alone cannot determine which retrieved memory would change agent behavior if neutralized; (2) Counterfactual attribution is prohibitively expensive for online detection, as it requires replaying downstream agent and tool execution for each candidate memory, exceeding real-time serving budgets. To address these challenges, we propose DriftGuard, a two-stage framework that decouples offline causal supervision from online estimation. Specifically, DriftGuard first constructs causal supervision through repeated paired counterfactual replays, continuing the original and single-memory neutralized trajectory suffixes from a shared pre-action state. The resulting annotations are aggregated by component-wise majority voting into a memory drift score (MDS) that captures action change, attack removal, violation removal, and task recovery. Guided by this supervision, DriftGuard then learns an online risk model that estimates MDS from pre-action information and flags drift-inducing memories via its maximum component risk, without online counterfactual rollouts. Extensive experiments demonstrate that DriftGuard accurately identifies drift-inducing memories and enables efficient online defense. Code is available at https://anonymous.4open.science/r/DriftGuard-7F0B.

## Duopra: From Full-Duplex Conversation to Full-Duplex Agency

DevFeed: [Duopra: From Full-Duplex Conversation to Full-Duplex Agency](<https://devfeed.tech/articles/duopra-from-full-duplex-conversation-to-full-duplex-agency-145586.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25773>)

Author: Chiyu Chen, Pengzhou Cheng, Yuezhao Ma, Xiaocheng Li, Can Liu, Haonan Hu, Yao Yao, Yifei Yang, Yijie Lu, Taolei Wang, Yansi Li, Muhuan Wu, Chenglu Sun, Gongshen Liu, Zhuosheng Zhang

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [function calling](<https://devfeed.tech/topics/function-calling.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>)

Tags: [audio](<https://devfeed.tech/tags/audio.md>), [backend](<https://devfeed.tech/tags/backend.md>), [bench](<https://devfeed.tech/tags/bench.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [execution](<https://devfeed.tech/tags/execution.md>), [frontend](<https://devfeed.tech/tags/frontend.md>), [full-duplex-bench](<https://devfeed.tech/tags/full-duplex-bench.md>), [interaction](<https://devfeed.tech/tags/interaction.md>), [interface](<https://devfeed.tech/tags/interface.md>), [latency](<https://devfeed.tech/tags/latency.md>), [low-latency](<https://devfeed.tech/tags/low-latency.md>), [model](<https://devfeed.tech/tags/model.md>), [natural-language](<https://devfeed.tech/tags/natural-language.md>), [pass](<https://devfeed.tech/tags/pass.md>), [pipeline](<https://devfeed.tech/tags/pipeline.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [realtime](<https://devfeed.tech/tags/realtime.md>), [repair](<https://devfeed.tech/tags/repair.md>), [request](<https://devfeed.tech/tags/request.md>), [requires](<https://devfeed.tech/tags/requires.md>), [v3](<https://devfeed.tech/tags/v3.md>)

### AI overview

Duopra addresses a gap in full-duplex agents: fluent turn-taking does not ensure timely, complete tool requests or corrections. It treats the backend as a tool that receives a self-contained natural-language request, using one action interface for direct calls and backend delegation. Training combines protocol SFT, full-duplex reinforcement learning for timely and complete execution, and projector-only repair of talker conditioning after thinker updates. On FDB v3, the authors report Pass@1 rising from 4.7% to 44.7% with the same backend; Duopra reaches 47.7% with direct tool calls (26.0% after SFT) and 49.3% with a DeepSeek backend. It also improves three FDB v1.0 takeover rates over the base model. The excerpt supports the authors' claim that task agency can coexist with key conversational agency.

### Source excerpt

arXiv:2611.25773v1 Announce Type: new Abstract: Full-duplex interaction models exhibit strong conversational agency, enabling natural, low-latency turn-taking, backchannels, and interruptions. However, completing tasks further requires task agency: timely tool invocation and argument revision following user corrections. Frontend-backend agents separate initiation from execution: the frontend decides whether and what to request, while the backend determines how. Execution can be offloaded, but unsent requests cannot be executed, and incomplete ones can misdirect the backend. On Full-Duplex-Bench v3 (FDB v3), most failures of Realtime-Venus-Audio + Harness in our streaming pipeline occur before a domain tool executes. We introduce Duopra, a full-duplex interaction model that learns task execution alongside conversational control. Duopra treats the backend as a tool, accepting a self-contained natural-language request. Direct calls and backend delegation share an action interface and requirements for faithful initiation and correct execution. Its three-stage training combines protocol SFT, full-duplex RL rewarding timely and complete execution, and projector-only repair of talker conditioning after thinker updates. With the same backend, Pass@1 rises from 4.7% to 44.7%; Duopra achieves 47.7% calling tools directly (26.0% after SFT) and 49.3% with a DeepSeek backend. It improves three FDB v1.0 takeover rates over the base model. These results show that task agency can coexist with key conversational agency.

## Spring AI 2.1.0 GA and 2.0.3 GA Available Now

DevFeed: [Spring AI 2.1.0 GA and 2.0.3 GA Available Now](<https://devfeed.tech/articles/spring-ai-2-1-0-ga-and-2-0-3-ga-available-now-146701.md>)

Original publisher: [Read original article](<https://spring.io/blog/2026/11/20/spring-ai-2-1-0-GA-available-now>)

Author: tzolov

Published: 2026-11-20T00:00:00Z

Content type: release

Language: en

Sources: [Spring](<https://devfeed.tech/sources/spring.md>)

Topics: [Spring AI](<https://devfeed.tech/topics/spring-ai.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [api](<https://devfeed.tech/tags/api.md>), [batch](<https://devfeed.tech/tags/batch.md>), [boot](<https://devfeed.tech/tags/boot.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [decision](<https://devfeed.tech/tags/decision.md>), [decision-models](<https://devfeed.tech/tags/decision-models.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [event-driven](<https://devfeed.tech/tags/event-driven.md>), [guardrail](<https://devfeed.tech/tags/guardrail.md>), [harness](<https://devfeed.tech/tags/harness.md>), [java](<https://devfeed.tech/tags/java.md>), [maven](<https://devfeed.tech/tags/maven.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [memory](<https://devfeed.tech/tags/memory.md>), [microservices](<https://devfeed.tech/tags/microservices.md>), [model-context-protocol](<https://devfeed.tech/tags/model-context-protocol.md>), [openai](<https://devfeed.tech/tags/openai.md>), [parts](<https://devfeed.tech/tags/parts.md>), [reactive](<https://devfeed.tech/tags/reactive.md>), [recall](<https://devfeed.tech/tags/recall.md>), [release](<https://devfeed.tech/tags/release.md>), [responses-api](<https://devfeed.tech/tags/responses-api.md>), [security](<https://devfeed.tech/tags/security.md>), [serverless](<https://devfeed.tech/tags/serverless.md>), [session](<https://devfeed.tech/tags/session.md>), [spring](<https://devfeed.tech/tags/spring.md>), [spring-ai](<https://devfeed.tech/tags/spring-ai.md>), [web-applications](<https://devfeed.tech/tags/web-applications.md>)

### AI overview

Spring AI 2.1.0 GA, built on Spring Boot 4.2, adds a stateless MCP 2026-07-28 protocol, typed Decision Models, ordered message parts that preserve reasoning and provider-specific payloads, and an OpenAI Responses API chat model. Its agent building blocks include a Session API with compaction and event recall (ChatMemory is deprecated for removal in 3.0), per-tool approval pauses, configurable per-turn call limits, and marked guardrail refusals. Eleven vector stores can now write and search vectors computed elsewhere, with width checks and stable-ID replacement. The release also includes reference-documentation changes and security fixes; Spring AI 2.0.3 carries the fixes on the 2.0.x line. Upgrading to 2.1 generally involves updating Spring AI and Spring Boot together, but some security fixes change behavior, and applications upgrading from 2.0.2 or earlier are directed to review the 2.0.2 and 2.0.3 notes.

### Source excerpt

On behalf of the Spring AI team and everyone who has contributed, I'm happy to announce that Spring AI 2.1.0 has been released and is now available from Maven Central! Release notes | Upgrade notes | Reference documentation Spring AI 2.1 arrives five months after 2.0.0 GA. It is built on Spring Boot 4.2 and contains more than 430 commits from 86 contributors. This post covers everything that changed since 2.0.0: the features of the 2.1 milestones and release candidates, what was added after RC2, and the additions that already shipped in 2.0.1 and 2.0.2. The short version: MCP 2026-07-28. Servers and clients for the new, stateless revision of the Model Context Protocol, next to the existing ones. Decision Models. A new model API for typed yes/no, choice and score decisions, with three providers. A richer message model. Messages hold ordered parts, so reasoning, tool calls and provider-specific blocks survive a round trip. The OpenAI Responses API is supported next to Chat Completions. The building blocks of an agent harness. The Session API, with session memory, compaction and recall, replaces ChatMemory. Tool calls can wait for a person's approval and are limited per turn, and guardrail refusals are marked as such. Your own embeddings. Eleven vector stores can write and search with vectors you computed yourself. New reference documentation, organized around tasks, with a quickstart and tutorials. Security. Fixes for 13 CVEs, which Spring AI 2.0.3 has as well. Next: Spring AI Agent. A new project for composing and orchestrating agents, built on Spring AI 2.1, starts incubating in early December. See What's next. The following map shows the building blocks of Spring AI 2.1, with the community projects in blue. It comes from the new reference documentation, where the blocks link to their pages. MCP 2026-07-28 The 2026-07-28 revision is a redesign of the Model Context Protocol. It is stateless: there is no initialize handshake and no session, every request carries the p

## Multi-Turn Agentic Reinforcement Learning with Tunix and Gemma 4 on Google TPUs

DevFeed: [Multi-Turn Agentic Reinforcement Learning with Tunix and Gemma 4 on Google TPUs](<https://devfeed.tech/articles/multi-turn-agentic-reinforcement-learning-with-tunix-and-gemma-4-on-google-tpus-144221.md>)

Original publisher: [Read original article](<https://developers.googleblog.com/multi-turn-agentic-reinforcement-learning-with-tunix-and-gemma-4-on-google-tpus/>)

Author: Wei Wei

Published: 2026-11-19T19:49:56.121412Z

Content type: tutorial

Language: en

Sources: [Google Developers Blog](<https://devfeed.tech/sources/google-developers-blog.md>)

Topics: [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [grpo](<https://devfeed.tech/topics/grpo.md>), [gemma](<https://devfeed.tech/topics/gemma.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [asynchronous](<https://devfeed.tech/tags/asynchronous.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cloud-tpu](<https://devfeed.tech/tags/cloud-tpu.md>), [gemma](<https://devfeed.tech/tags/gemma.md>), [gemma-4](<https://devfeed.tech/tags/gemma-4.md>), [google](<https://devfeed.tech/tags/google.md>), [group-relative-policy-optimization](<https://devfeed.tech/tags/group-relative-policy-optimization.md>), [grpo](<https://devfeed.tech/tags/grpo.md>), [learning](<https://devfeed.tech/tags/learning.md>), [llm](<https://devfeed.tech/tags/llm.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [tool](<https://devfeed.tech/tags/tool.md>), [tpus](<https://devfeed.tech/tags/tpus.md>), [training](<https://devfeed.tech/tags/training.md>), [using](<https://devfeed.tech/tags/using.md>), [vllm](<https://devfeed.tech/tags/vllm.md>)

### AI overview

The post describes training Gemma 4 E2B as a multi-turn Plane Strike agent with Tunix agentic GRPO on one Cloud TPU v6e-8. The game requires repeated board observations and `strike` tool calls; a custom environment, Gemma 4 chat parser, and conversation agent handle state, native tool responses, and action validation. Asynchronous vLLM rollouts with prefix caching support variable-length turns, while exact token continuity avoids retokenization drift. Starting from a 9.4% held-out win rate and 22.44 average strikes, the model reached 95% wins and about 16 strikes after 100 training steps; tool-call rate approached 100% and invalid actions approached zero. All strikes currently share a trajectory-level reward, which the post identifies as a limitation that finer per-turn credit assignment could address.

### Source excerpt

This post walks developers through training a multi-turn, tool-using LLM agent using Tunix and Gemma 4 on Cloud TPUs. By leveraging agentic Group Relative Policy Optimization (GRPO) and asynchronous vLLM rollouts, the approach significantly improves complex reasoning and tool compliance in interactive environments.

## What Is Tool Calling? How LLMs and AI Agents Take Action

DevFeed: [What Is Tool Calling? How LLMs and AI Agents Take Action](<https://devfeed.tech/articles/what-is-tool-calling-how-llms-and-ai-agents-take-action-145886.md>)

Original publisher: [Read original article](<https://montecarlo.ai/blog-what-is-tool-calling>)

Author: Virna Sekuj

Published: 2026-11-19T16:58:24Z

Content type: article

Language: en

Sources: [Monte Carlo](<https://devfeed.tech/sources/monte-carlo.md>)

Topics: [function calling](<https://devfeed.tech/topics/function-calling.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [LLM observability](<https://devfeed.tech/topics/llm-observability.md>), [API](<https://devfeed.tech/topics/api.md>), [Tool](<https://devfeed.tech/topics/tool.md>)

Tags: [2023](<https://devfeed.tech/tags/2023.md>), [2026](<https://devfeed.tech/tags/2026.md>), [actions](<https://devfeed.tech/tags/actions.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-behavior](<https://devfeed.tech/tags/agent-behavior.md>), [agent-frameworks](<https://devfeed.tech/tags/agent-frameworks.md>), [agent-trust](<https://devfeed.tech/tags/agent-trust.md>), [agent-trust-glossary](<https://devfeed.tech/tags/agent-trust-glossary.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [api](<https://devfeed.tech/tags/api.md>), [database](<https://devfeed.tech/tags/database.md>), [function](<https://devfeed.tech/tags/function.md>), [function-calling](<https://devfeed.tech/tags/function-calling.md>), [glossary](<https://devfeed.tech/tags/glossary.md>), [large-language-model](<https://devfeed.tech/tags/large-language-model.md>), [llm-tool-calling](<https://devfeed.tech/tags/llm-tool-calling.md>), [llms](<https://devfeed.tech/tags/llms.md>), [querying](<https://devfeed.tech/tags/querying.md>), [searching](<https://devfeed.tech/tags/searching.md>), [structured](<https://devfeed.tech/tags/structured.md>), [tool-calling](<https://devfeed.tech/tags/tool-calling.md>), [tool-calling-examples](<https://devfeed.tech/tags/tool-calling-examples.md>), [tool-calling-vs-mcp](<https://devfeed.tech/tags/tool-calling-vs-mcp.md>), [tool-use-ai-agents](<https://devfeed.tech/tags/tool-use-ai-agents.md>), [uses](<https://devfeed.tech/tags/uses.md>)

### AI overview

Tool calling lets an LLM request an action by returning a structured tool name and arguments; the application validates and executes the request, sends back the result, and the model uses it to answer or make another call. Function calling and tool use are alternate names for the same mechanism. MCP standardizes how compatible applications discover and connect to tools; it builds on tool calling rather than replacing it. Calls can fail through wrong tool choice, bad arguments, skipped checks, retries, mishandled errors, changed responses, or unintended access, often without obvious errors. The text recommends tracing each call and its result, validating arguments and permissions before execution, watching for unexpected patterns and loops, tracking cost and latency, monitoring the underlying data, and limiting or requiring approval for high-impact actions.

### Source excerpt

Tool calling, also called function calling, is how a large language model asks an application to run a function, such as querying a database, calling an API, or searching documents, and then uses the result in its answer. The model doesn't run the tool itself. It produces a structured request naming the tool and its arguments, the application executes it, and the result goes back to the model.

## Build Tool-Calling Parsers in Rust with AI Agents

DevFeed: [Build Tool-Calling Parsers in Rust with AI Agents](<https://devfeed.tech/articles/build-tool-calling-parsers-in-rust-with-ai-agents-143887.md>)

Original publisher: [Read original article](<https://developer.nvidia.com/blog/build-tool-calling-parsers-in-rust-with-ai-agents/>)

Author: Keiven Chang

Published: 2026-11-19T16:30:46Z

Content type: tutorial

Language: en

Sources: [NVIDIA Developer](<https://devfeed.tech/sources/nvidia-developer.md>), [NVIDIA Technical Blog](<https://devfeed.tech/sources/nvidia-technical-blog.md>)

Topics: [function calling](<https://devfeed.tech/topics/function-calling.md>), [rust-ai](<https://devfeed.tech/topics/rust-ai.md>), [Test-driven development](<https://devfeed.tech/topics/tdd.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [Nemotron](<https://devfeed.tech/topics/nemotron.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [3](<https://devfeed.tech/tags/3.md>), [adapter](<https://devfeed.tech/tags/adapter.md>), [adapting](<https://devfeed.tech/tags/adapting.md>), [agentic-ai-generative-ai](<https://devfeed.tech/tags/agentic-ai-generative-ai.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [argument](<https://devfeed.tech/tags/argument.md>), [build](<https://devfeed.tech/tags/build.md>), [call](<https://devfeed.tech/tags/call.md>), [can](<https://devfeed.tech/tags/can.md>), [developer-tools-techniques](<https://devfeed.tech/tags/developer-tools-techniques.md>), [dynamo](<https://devfeed.tech/tags/dynamo.md>), [frontend](<https://devfeed.tech/tags/frontend.md>), [inference-performance](<https://devfeed.tech/tags/inference-performance.md>), [leak](<https://devfeed.tech/tags/leak.md>), [markup](<https://devfeed.tech/tags/markup.md>), [model](<https://devfeed.tech/tags/model.md>), [rust](<https://devfeed.tech/tags/rust.md>), [tool](<https://devfeed.tech/tags/tool.md>), [tool-calling](<https://devfeed.tech/tags/tool-calling.md>), [turns](<https://devfeed.tech/tags/turns.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

Tool-call parsers must separate text, reasoning, and calls while preserving streaming boundaries and avoiding lost data, duplicate calls, or leaked model markup. The article's Rust workflow uses frontend-crates and its shared Conformance Corpus: adapt applicable UNIFIED fixtures, create and human-review GOLDEN expectations against the model specification and related sources, then implement the parser and inspect captured results in the local conformance report. An unavailable tool is reported as generated; the serving layer checks availability, authorization, schema, and policy. Incomplete calls require explicit buffering and end-of-stream recovery rules rather than guessing missing content. CPU fixtures suffice for parser conformance, while end-to-end validation needs Dynamo runtime, discovery services, and model hardware. Further checks should measure speed and memory across input and chunk sizes, verify incremental event delivery and truncation behavior, and test the configured quote-handling policy. Guided decoding may reduce malformed calls but cannot prevent truncation or incorrect tool choices.

### Source excerpt

A tool call can arrive in pieces, an argument can be cut short, or model-specific markup can leak into the answer. A frontend parser turns that output, received...

## Bad Genius: Counterfactual-Guided Harness Evolution Beyond Task-Specific Shortcuts

DevFeed: [Bad Genius: Counterfactual-Guided Harness Evolution Beyond Task-Specific Shortcuts](<https://devfeed.tech/articles/bad-genius-counterfactual-guided-harness-evolution-beyond-task-specific-shortcuts-142560.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2609.18366>)

Author: Guojun Zhu, Xunheng Huang, Peng Yin, Jiahui Xie, Sanguo Zhang, Doudou Zhou

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [benchmark overfitting machine learning](<https://devfeed.tech/topics/benchmark-overfitting-machine-learning.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agent-harness](<https://devfeed.tech/tags/agent-harness.md>), [archive](<https://devfeed.tech/tags/archive.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [evolution](<https://devfeed.tech/tags/evolution.md>), [formatting](<https://devfeed.tech/tags/formatting.md>), [foundation-model](<https://devfeed.tech/tags/foundation-model.md>), [harness](<https://devfeed.tech/tags/harness.md>), [order](<https://devfeed.tech/tags/order.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [search](<https://devfeed.tech/tags/search.md>), [semantics](<https://devfeed.tech/tags/semantics.md>), [shortcuts](<https://devfeed.tech/tags/shortcuts.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>), [swe-bench](<https://devfeed.tech/tags/swe-bench.md>), [swe-bench-pro](<https://devfeed.tech/tags/swe-bench-pro.md>)

### AI overview

CHASE tests whether agent-harness improvements survive protocol changes that preserve task semantics. A Proposer edits a harness around a fixed foundation model; a Challenger searches for changes that reduce its gain over the initial harness. Executable checks verify task preservation, and confirmed held-out gain losses are archived to constrain later updates. The authors state conditions under which a finite archive bounds losses on an ideal benchmark without shortcuts. On Syn-Ledger with known shortcuts removed, CHASE reached 38.30% accuracy versus 18.27% for HarnessEvolve. On 21 SWE-Bench Pro V2 HARD-51 certification tasks, changing issue formatting and tool-output field order cut HarnessEvolve's gain from 8.57 to 3.81 percentage points; CHASE retained 10.48 of its 11.43-point gain.

### Source excerpt

arXiv:2609.18366v4 Announce Type: replace Abstract: An optimized agent harness can improve on held-out tasks by relying on formatting conventions that those tasks share with the rest of the benchmark. We introduce Counterfactual Harness Search and Evolution (CHASE) to test whether harness improvements survive changes to the evaluation protocol. A Proposer edits the harness around a fixed foundation model. A Challenger then searches for a protocol change that preserves the tasks but reduces the updated harness's gain over the initial harness. Executable checks test whether the transformation preserves task semantics. If the gain loss is confirmed on held-out tasks, CHASE adds the transformation to an archive that constrains subsequent harness updates. We give conditions under which this finite archive bounds the gain loss on an ideal benchmark with shortcuts removed. We evaluate CHASE on Syn-Ledger, OfficeQA, FinanceBench, and SWE-Bench Pro V2. On Syn-Ledger, removing the known shortcuts yields 38.30% accuracy for CHASE and 18.27% for HarnessEvolve. On the 21 certification tasks from SWE-Bench Pro V2 HARD-51, changing the issue formatting and tool-output field order reduces HarnessEvolve's gain over the initial harness from 8.57 to 3.81 percentage points. The final CHASE harness retains 10.48 points of its 11.43-point gain under the same change.

## CADFather: Autonomous CAD Reconstruction through Coordinated Tool Use

DevFeed: [CADFather: Autonomous CAD Reconstruction through Coordinated Tool Use](<https://devfeed.tech/articles/cadfather-autonomous-cad-reconstruction-through-coordinated-tool-use-142573.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.09127>)

Author: Gennadiy Savrasov, Maksim Elistratov, Nikita Gavrilov, Albert Garifullin, Oleg Pavlov, Soslan Kabisov, Vladimir Frolov, Anton Konushin, Andrey Kuznetsov, Dmitrii Zhemchuzhnikov

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Agentic AI](<https://devfeed.tech/topics/what-is-agentic-ai.md>), [agentic workflows](<https://devfeed.tech/topics/agentic-workflows.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [3D](<https://devfeed.tech/topics/3d.md>), [Tool](<https://devfeed.tech/topics/tool.md>)

Tags: [3d](<https://devfeed.tech/tags/3d.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [autonomous](<https://devfeed.tech/tags/autonomous.md>), [cad](<https://devfeed.tech/tags/cad.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [tool-use](<https://devfeed.tech/tags/tool-use.md>)

### AI overview

CADFather reconstructs editable parametric CAD programs from 3D meshes by coordinating complementary proposal tools with numerical refinement. A vision-language assistant inspects target and intermediate renders, selects candidate programs and tools, controls proposal counts, and decides when to stop. Proposed or refined programs are executed and evaluated for feedback; the system keeps alternative candidates and preserves the best valid result. It uses pretrained models without additional training. The abstract reports evaluation of reconstruction quality and execution validity on DeepCAD, Fusion360, MCB, CADENA-Bench, CADBench, and BenchCAD, plus analysis of computational cost and its trade-off with quality, but gives no comparative results.

### Source excerpt

arXiv:2610.09127v2 Announce Type: replace Abstract: Reconstructing an editable CAD model from a 3D shape remains a challenging engineering task. Existing methods can propose CAD operations, but no single source of proposals works equally well across different part geometries and stages of reconstruction. We introduce CADFather, an autonomous agentic system that coordinates complementary tools to recover parametric CAD programs from 3D meshes. A vision-language assistant inspects renders of the target and intermediate reconstructions, then decides which candidate CAD programs to extend, which tools to invoke, how many proposals to generate, and when to finish. Learned and algorithmic tools propose CAD operations, while numerical optimization refines the parameters of existing programs. Proposed or refined programs are executed and evaluated to provide feedback for subsequent decisions. The agent maintains alternative candidate programs for each target part and preserves the best valid result throughout reconstruction. CADFather uses pretrained generation and assistant models without additional training. We evaluate reconstruction quality and execution validity on the full DeepCAD, Fusion360, and MCB test sets, as well as on CADENA-Bench, CADBench, and BenchCAD. We additionally analyze computational cost and the trade-off between cost and reconstruction quality.

## MedExAgent: Training LLM Agents to Ask, Examine, and Diagnose in Noisy Clinical Environments

DevFeed: [MedExAgent: Training LLM Agents to Ask, Examine, and Diagnose in Noisy Clinical Environments](<https://devfeed.tech/articles/medexagent-training-llm-agents-to-ask-examine-and-diagnose-in-noisy-clinical-environments-142519.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2605.07058>)

Author: Yicheng Gao, Xiaolin Zhou, Yahan Li, Yue Zhao, Ruishan Liu

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>)

Tags: [ablation](<https://devfeed.tech/tags/ablation.md>), [agents](<https://devfeed.tech/tags/agents.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [diagnosis](<https://devfeed.tech/tags/diagnosis.md>), [environment](<https://devfeed.tech/tags/environment.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [llms](<https://devfeed.tech/tags/llms.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

MedExAgent treats diagnosis as an interactive, uncertain process: an agent can question a patient, order medical exams as tool calls, or issue a diagnosis. Its environment models seven types of patient noise and three types of exam noise. Training first uses supervised fine-tuning on synthetic interviews structured around the Calgary-Cambridge model, then DAPO optimizes a combined reward for diagnostic accuracy, tool-call quality, and exam costs that include financial expense and patient discomfort. The abstract reports diagnostic performance comparable to larger models with cost-efficient exam strategies, but gives no specific metrics or details of the experiments.

### Source excerpt

arXiv:2605.07058v2 Announce Type: replace-cross Abstract: Real-world clinical diagnosis is a complex process in which the doctor is required to obtain information from both interaction with the patient and conducting medical exams. Additionally, the doctor needs to adapt to different patient personas, as well as noisy and incomplete information that can happen at any time during the process. However, existing benchmarks for medical LLMs and methods for automatic diagnosis largely simplify this process by reducing it to single-turn question answering, noise-free conversations, or sequential exam making, etc., ignoring the interactive and uncertain nature of clinical diagnosis. In this paper, we aim to address this gap by formalizing clinical diagnosis as a Partially Observable Markov Decision Process (POMDP) with three action types: questioning the patient, ordering medical exams as tool calls, and issuing a diagnosis. We also introduce a systematic noise model comprising seven patient noise types and three exam noise types. Using our proposed environment, we train an effective diagnosis agent, \textbf{MedExAgent}, through a two-stage pipeline that first performs supervised finetuning on synthetic conversations structured after the Calgary-Cambridge model for clinical interviews, and then applies DAPO to optimize a composite reward capturing diagnostic accuracy, tool call quality, and exam cost including financial cost and patient discomfort. Through extensive experiments and ablation studies, we demonstrate that MedExAgent achieves diagnostic performance comparable to larger models while maintaining cost-efficient examination strategies.

## Worth Another Look:Constrained World-Model Planning for Visual Inspection

DevFeed: [Worth Another Look:Constrained World-Model Planning for Visual Inspection](<https://devfeed.tech/articles/worth-another-look-constrained-world-model-planning-for-visual-inspection-142764.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.24187>)

Author: Ziyue Kang, Zixi Qi, Yuchen Dai, Nana Lv, Xirong Cheng, Liang Liu, Ying Li, Haoyan Wei, Haiyang Yao

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [World models](<https://devfeed.tech/topics/world-models.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [Physical AI](<https://devfeed.tech/topics/physical-ai.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [beam-search](<https://devfeed.tech/tags/beam-search.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [budget](<https://devfeed.tech/tags/budget.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-mm](<https://devfeed.tech/tags/cs-mm.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [environmental-monitoring](<https://devfeed.tech/tags/environmental-monitoring.md>), [fallback](<https://devfeed.tech/tags/fallback.md>), [inspection](<https://devfeed.tech/tags/inspection.md>), [lightweight](<https://devfeed.tech/tags/lightweight.md>), [lookahead](<https://devfeed.tech/tags/lookahead.md>), [observability](<https://devfeed.tech/tags/observability.md>), [offline](<https://devfeed.tech/tags/offline.md>), [online](<https://devfeed.tech/tags/online.md>), [optical](<https://devfeed.tech/tags/optical.md>), [pairing](<https://devfeed.tech/tags/pairing.md>), [performance](<https://devfeed.tech/tags/performance.md>), [planning](<https://devfeed.tech/tags/planning.md>), [policies](<https://devfeed.tech/tags/policies.md>), [pricing](<https://devfeed.tech/tags/pricing.md>), [visual-inspection](<https://devfeed.tech/tags/visual-inspection.md>), [world-model](<https://devfeed.tech/tags/world-model.md>)

### AI overview

The paper proposes budget-aware visual inspection that combines a compact recurrent world model with exact symbolic feasibility and tariff rules. Receding-horizon beam search replans schedules for class queries, crops, image enhancements, and stopping; slack-aware shadow pricing and a legal fallback support larger budgets. Across eight terrestrial, optical, and sonar benchmarks, multi-step lookahead beats static schedules and search-free offline RL under tight budgets, with gains of up to 28.4 AP points over static priority on RUOD and 15.1 on ARIS. Dynamic pricing improves large-budget results across all eight domains, at an 11 ms planning cost. The approach uses frozen detectors without retraining.

### Source excerpt

arXiv:2611.24187v1 Announce Type: cross Abstract: Visual inspection in field robotics and environmental monitoring must localize multiple targets under strict compute and sensing caps. Exhaustive sweeps waste calls on absent classes, while myopic and search-free policies stumble over heterogeneous tariff cliffs or discard viewport edits whose value emerges only after re-querying. We formulate inspection over frozen detectors as neuro-symbolic observation planning under partial observability, decoupling a compact recurrent world model (WM) from exact symbolic feasibility and tariff rules. Receding-horizon beam search with online replanning jointly schedules class queries, crops, enhancements, and termination (stop), paired with slack-aware shadow pricing and legal fallback at larger budgets. Across eight terrestrial, optical, and sonar benchmarks, multi-step lookahead outperforms static schedules and search-free offline RL under tight budgets -- gaining up to 28.4 AP percentage points over static priority on RUOD and 15.1 on ARIS -- while dynamic pricing lifts large-budget performance on all eight domains at 11 ms planning cost. Ultimately, pairing lightweight learned dynamics with exact symbolic constraints turns frozen detectors into adaptive, budget-aware inspection agents without retraining.

## Risk-Triggered Search: Behavior-Preserving Safe Control

DevFeed: [Risk-Triggered Search: Behavior-Preserving Safe Control](<https://devfeed.tech/articles/risk-triggered-search-behavior-preserving-safe-control-142792.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.24409>)

Author: Xinyu Deng, Bohao Lan, Kunpeng Li, Mingkai Yuan, Weihang Pan, Binbin Lin

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [safety mechanism](<https://devfeed.tech/topics/safety-mechanism.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [VAE](<https://devfeed.tech/topics/vae.md>)

Tags: [benchmark](<https://devfeed.tech/tags/benchmark.md>), [control](<https://devfeed.tech/tags/control.md>), [controller](<https://devfeed.tech/tags/controller.md>), [cross-entropy-method](<https://devfeed.tech/tags/cross-entropy-method.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [embedding](<https://devfeed.tech/tags/embedding.md>), [navigation](<https://devfeed.tech/tags/navigation.md>), [risk](<https://devfeed.tech/tags/risk.md>), [search](<https://devfeed.tech/tags/search.md>), [world-models](<https://devfeed.tech/tags/world-models.md>)

### AI overview

Risk-Triggered Search (RTS) aims to prevent process violations in frozen JEPA visual world models while preserving safe nominal behavior. It predicts risk by unrolling action-conditioned latent prefixes and training a lightweight risk head with process-violation supervision. Safe plans execute directly; alarms trigger local Cross-Entropy Method search constrained by the same risk criterion. An amplitude-preserving metric is designed to retain tangential progress and avoid stalling. In the reported five-task benchmark, RTS improved Safe Arrival Rate over Direct by 12.67-41.67 percentage points on navigation and articulated-control tasks and 2.08/4.17 points on Cube/PushT boundary cohorts. Across 1,968 paired trajectories, it matched every-step search in action identity and reduced controller time by 17.71%-70.41%.

### Source excerpt

arXiv:2611.24409v1 Announce Type: new Abstract: Pre-trained visual world models generate multi-step action plans from current and goal observations, yet terminal-proximity objectives fail to prevent forbidden-zone traversal or rigid-body collision sliding. Purely visual interfaces may lack analytical dynamics, structured states, or explicit collision geometry, requiring internal prediction of action-conditioned process risk. We propose Risk-Triggered Search (RTS) for frozen Joint-Embedding Predictive Architecture (JEPA) world models. Rather than relying on static state features that cannot distinguish candidate plans sharing an observation, RTS unrolls action-conditioned future latent prefixes in the frozen world model and trains a lightweight risk head with process-violation supervision. To intervene selectively while preserving safe nominal behavior, RTS uses the same predicted prefix-risk criterion as both the search trigger gate and hard feasibility gate: safe nominal plans execute directly, whereas alarms trigger local Cross-Entropy Method (CEM) search to minimize nominal deviation within the feasible set. To prevent isotropic Euclidean minimal modification from shrinking obstacle-directed actions into zero-amplitude stalling, we introduce an amplitude-preserving metric $d_{\mathrm{AP}}$ that retains tangential progression while maintaining a unique zero at the nominal plan. Under unified gating and this unique-zero condition, bypassing safe-state search yields closed-loop trajectories strictly identical to every-step same-kernel search. Evaluated under a process-safety benchmark and Safe Arrival Rate (SAR) metric across five tasks, RTS improves SAR over Direct by 12.67--41.67 percentage points on navigation and articulated control and +2.08/+4.17 pp on Cube/PushT boundary cohorts, maintaining 100% action identity with every-step search while cutting controller time by 17.71%--70.41% across 1,968 paired trajectories.

## Space-X: Self-Evolving Spatial Agents Weaving Clues into Evidence Chains

DevFeed: [Space-X: Self-Evolving Spatial Agents Weaving Clues into Evidence Chains](<https://devfeed.tech/articles/space-x-self-evolving-spatial-agents-weaving-clues-into-evidence-chains-142740.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.24004>)

Author: Zijian Zhang, Wenqing Tian, Qiang Liu, Liang Wang

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [capability](<https://devfeed.tech/tags/capability.md>), [complete](<https://devfeed.tech/tags/complete.md>), [continuous](<https://devfeed.tech/tags/continuous.md>), [control](<https://devfeed.tech/tags/control.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [decision-rules](<https://devfeed.tech/tags/decision-rules.md>), [execution](<https://devfeed.tech/tags/execution.md>), [frame](<https://devfeed.tech/tags/frame.md>), [framework](<https://devfeed.tech/tags/framework.md>), [gemini-3-8-flash](<https://devfeed.tech/tags/gemini-3-8-flash.md>), [gpt-6-sol](<https://devfeed.tech/tags/gpt-6-sol.md>), [interfaces](<https://devfeed.tech/tags/interfaces.md>), [memory](<https://devfeed.tech/tags/memory.md>), [models](<https://devfeed.tech/tags/models.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [reference](<https://devfeed.tech/tags/reference.md>), [spatial](<https://devfeed.tech/tags/spatial.md>), [spatial-reasoning](<https://devfeed.tech/tags/spatial-reasoning.md>), [targets](<https://devfeed.tech/tags/targets.md>)

### AI overview

Space-X uses two agents to build traceable evidence chains for spatial reasoning. An Execution Agent gathers, associates, and evaluates evidence for each problem using role-specific memory; a Control Agent examines execution traces to find breaks and revises tools, interfaces, skills, memory organization, and decision rules. Evidence needs guide capability development, while usable results guide actions. The abstract reports gains over baselines across multiple benchmarks and four base models; for GPT-6-sol, answer accuracy rises 50.39% on ViewSpatial and 23.07% on MMSI, while Gemini-3.8-Flash gains 47.50% and 25.22% on those benchmarks, respectively.

### Source excerpt

arXiv:2611.24004v1 Announce Type: new Abstract: Spatial reasoning connects visual observations, geometric estimates, and linguistic decisions into a coherent derivation. Yet tool-augmented agents often acquire evidence without clear targets, combine outputs under inconsistent object or reference-frame semantics, and misjudge evidence sufficiency. Thus, more tools or calls do not reliably improve reasoning. We introduce Space-X, a spatial-reasoning framework that self-evolves toward complete, continuous evidence-chain construction. An Execution Agent acquires, associates, and judges evidence per problem using role-specific memory. A Control Agent diagnoses evidence-chain breaks from execution traces and jointly revises tools, interfaces, skills, memory organization, and decision rules. This loop co-evolves capability and evidence use: evidence needs guide capability development; usable results guide actions. Across multiple benchmarks, Space-X yields gains with four base models. Relative to baselines, GPT-6-sol improves answer accuracy on ViewSpatial and MMSI by 50.39% and 23.07%, respectively; Gemini-3.8-Flash gains 47.50% and 25.22% on these benchmarks. This provides a unified route from fragmented spatial clues to traceable, derivable answers.

## Intent Contracts: Taking Tools Away from Agents

DevFeed: [Intent Contracts: Taking Tools Away from Agents](<https://devfeed.tech/articles/intent-contracts-taking-tools-away-from-agents-142742.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.24025>)

Author: Ziyuan Wang, Genliang Zhu, Xiaobo Chen, Chu Wang

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [LLM security](<https://devfeed.tech/topics/llm-security.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [authorization](<https://devfeed.tech/tags/authorization.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [code](<https://devfeed.tech/tags/code.md>), [contracts](<https://devfeed.tech/tags/contracts.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [model](<https://devfeed.tech/tags/model.md>), [models](<https://devfeed.tech/tags/models.md>), [object](<https://devfeed.tech/tags/object.md>), [permission](<https://devfeed.tech/tags/permission.md>), [planner](<https://devfeed.tech/tags/planner.md>), [planning](<https://devfeed.tech/tags/planning.md>), [plugin](<https://devfeed.tech/tags/plugin.md>), [review](<https://devfeed.tech/tags/review.md>), [run](<https://devfeed.tech/tags/run.md>), [scope](<https://devfeed.tech/tags/scope.md>), [state](<https://devfeed.tech/tags/state.md>), [target](<https://devfeed.tech/tags/target.md>), [tool](<https://devfeed.tech/tags/tool.md>), [tools](<https://devfeed.tech/tags/tools.md>), [using](<https://devfeed.tech/tags/using.md>)

### AI overview

Intent contracts keep tools away from the planning model: for each step it submits a logged goal, a verbatim user quotation, and scope fields; code validates these, and a trusted model checks intent and writes the tool call. In three AmPermBench task families using gpt-6.1-terra and GLM-5.3-Flash, every changed object under contracts appeared in the producing contract's goal. When the planner added unrequested work, contracts completed the user's task in 27/30 runs, versus 16/30 for a reviewer that could only refuse whole commands. Contracts did not reduce unauthorized actions and were indistinguishable from blind review. After environment terms were clarified, state-changing conditions targeted the wrong object in 40-49% of runs; the trusted model could not check the planner's choice of object for ambiguous references. The evaluation covers benign over-reach with one model pair, not adversarial planners.

### Source excerpt

arXiv:2611.24025v1 Announce Type: cross Abstract: Intent contracts are an authorization interface for tool-using LLM agents in which the planning model holds no tools. For each step it submits a short logged contract: a goal, a verbatim quotation of the user, and scope fields. Code validates the contract, and a small trusted model checks it against the user's words and writes the tool call itself. Provided the validation and dispatch code is correct, every executed call is written by the trusted model, the planner's influence on a call is confined to a contract logged before it, and the trusted side decides how much of a request to carry out. We compare contracts with two reviewers of planner-written commands: one sees only the user's words and the command, the information boundary of a reasoning-blind permission classifier, and one also sees the planner's goal and a checked quotation. With gpt-6.1-terra as planner and GLM-5.3-Flash as the trusted model on three AmPermBench task families, every object changed under contracts is named in the goal of the contract that produced the change; under review the planner names objects in its commands and rarely in its goals. When the planner adds work the user never asked for, contracts complete the user's task in 27/30 runs and the intent-visible reviewer in 16/30, because a reviewer can only refuse a whole command. Contracts do not reduce unauthorized actions. They are indistinguishable from blind review, and once the benchmark's environment terms are clarified, every condition that changes state hits a wrong target in 40-49% of those runs. That error is the planner's choice of objects for an ambiguous reference, which no trusted model here can see the environment to check; telling the executor to decline when unsure does not change it. We evaluate benign over-reach with one pair of models, not adversarial planners. We release the plugin and run records.

## Safe Personal RSI: Evaluating and Repairing Personal-Agent Updates

DevFeed: [Safe Personal RSI: Evaluating and Repairing Personal-Agent Updates](<https://devfeed.tech/articles/safe-personal-rsi-evaluating-and-repairing-personal-agent-updates-142190.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23651>)

Author: Jiapeng Pei, Yuhang Yao, Sitan Liu, Hanxin Chen, Hao Li, Guogeng Li, Jiapei Liu, Ruijie Chen, Hongshan Lin, Yang Mo, Tianyu Shi, Shangding Gu

Published: 2026-11-19T05:00:00Z

Content type: news

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Software Engineering](<https://devfeed.tech/sources/arxiv-software-engineering.md>)

Topics: [ai alignment](<https://devfeed.tech/topics/ai-alignment.md>), [AI Agent](<https://devfeed.tech/topics/ai-agent.md>), [AI Prompt Templates](<https://devfeed.tech/topics/ai-prompt-templates.md>), [gaia](<https://devfeed.tech/topics/gaia.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [ai-governance](<https://devfeed.tech/topics/ai-governance.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [actions](<https://devfeed.tech/tags/actions.md>), [adoption](<https://devfeed.tech/tags/adoption.md>), [agent](<https://devfeed.tech/tags/agent.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [attack](<https://devfeed.tech/tags/attack.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [authorization](<https://devfeed.tech/tags/authorization.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [calendar](<https://devfeed.tech/tags/calendar.md>), [can](<https://devfeed.tech/tags/can.md>), [cases](<https://devfeed.tech/tags/cases.md>), [changes](<https://devfeed.tech/tags/changes.md>), [code](<https://devfeed.tech/tags/code.md>), [completion](<https://devfeed.tech/tags/completion.md>), [component](<https://devfeed.tech/tags/component.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-se](<https://devfeed.tech/tags/cs-se.md>), [personal](<https://devfeed.tech/tags/personal.md>), [personal-agent](<https://devfeed.tech/tags/personal-agent.md>), [rsi](<https://devfeed.tech/tags/rsi.md>), [updates](<https://devfeed.tech/tags/updates.md>)

### AI overview

Safe Personal RSI evaluates proposed changes to Kicker's reusable instructions while keeping model weights and runtime code fixed. It tests current and proposed prompts, repeats failed cases, temporarily reverses changes to guide repairs, and adopts an update only if it passes safety and task-performance checks; otherwise, the current instructions remain. In a 29-case benchmark across three models and two risky edits, it rejected 20 of 21 update attempts. A confirmed GPT-5.6 Luna repair reduced attack violations from 10.7% to 0% and raised safe task completion from 89.7% to 100%, matching a broad handwritten authority rule. A repair transferred from Qwen2.5-7B-Instruct to Qwen3-14B reduced violations under stronger attacks from 69.0% to 17.9% and raised safe completion from 19.5% to 64.4%. Separate runtime probes found missing local OAuth scope checks in the evaluated snapshot.

### Source excerpt

arXiv:2611.23651v1 Announce Type: new Abstract: Kicker is a personal health assistant. Changes to an assistant's reusable instructions can improve task handling while changing what it discloses, which requests it follows, and how it reports actions. We study Safe Personal RSI, an update-evaluation component developed for Kicker. We restrict recursive self-improvement (RSI) to prompt updates, with model weights and runtime code fixed. The procedure tests proposed and current instructions, repeats failing cases, temporarily undoes changes to guide repairs, and checks safety and task performance before adoption. If no version passes, the current instructions stay in use. Our executable benchmark contains 29 email, calendar, health, and marketplace cases, with separate probes for authorization reuse, memory, and tool-output injection. Across three models and two deliberately risky edits, the procedure rejects 20 of 21 update attempts. In repair confirmation, an admitted GPT-5.6 Luna repair reduces attack violations from 10.7% to 0% and increases safe task completion from 89.7% to 100%; a broad handwritten authority rule matches these outcomes. Under stronger attacks, a repair transferred from Qwen2.5-7B-Instruct to Qwen3-14B reduces violations from 69.0% to 17.9% and increases safe task completion from 19.5% to 64.4%. Separate runtime probes examine memory-entry deletion and identify missing local OAuth scope checks in the evaluated snapshot.

## Thinking Fast, Anywhere: One System-1 for the Decision Points of Any Agent Workflow

DevFeed: [Thinking Fast, Anywhere: One System-1 for the Decision Points of Any Agent Workflow](<https://devfeed.tech/articles/thinking-fast-anywhere-one-system-1-for-the-decision-points-of-any-agent-workflow-142701.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23751>)

Author: Rui Li, Liyang He, Zheng Zhang, Zhenya Huang, Qi Liu

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [agentic workflows](<https://devfeed.tech/topics/agentic-workflows.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Benchmark](<https://devfeed.tech/topics/benchmark.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [decision](<https://devfeed.tech/tags/decision.md>), [latency](<https://devfeed.tech/tags/latency.md>), [performance](<https://devfeed.tech/tags/performance.md>), [workflow](<https://devfeed.tech/tags/workflow.md>)

### AI overview

AnyS1 uses procedural skills to identify enumerable decision points across workflows, avoiding the need for environment-specific hooks. Offline, it mines decisions from over 1,000 public skills, gathers execution-grounded states in isolated containers to build Any-DP, and trains a calibrated 4B pointer System-1. At runtime, routine choices are made in one forward pass; decisions below a confidence threshold fall back to the larger System-2 model. On three unseen benchmarks and across distinct System-2 model families, the authors report preserved task reliability, 75-91% lower episode cost, and 63-92% less model-call time than an uncompiled System-2 agent, with performance maintained under zero-shot threshold transfer.

### Source excerpt

arXiv:2611.23751v1 Announce Type: new Abstract: Autonomous LLM agents incur substantial cost and latency by invoking a large model (System-2) at every step, even when many intermediate actions choose among a small set of known candidates. Existing dual-system agents offload such decisions to a lightweight System-1 model, but rely on workflow-specific hooks authored per environment. We present AnyS1, a framework that turns human-authored procedural skills, whose instructions naturally specify where choices arise and which options apply, into a shared, domain-general source of decision points. Offline, AnyS1 mines grounded enumerable decisions across over 1,000 public skills, gathers execution-grounded trajectory states inside isolated containers to construct Any-DP, and trains a single calibrated 4B pointer System-1 with label smoothing. At run time, compiled workflows resolve routine decisions in one forward pass and fall back to System-2 when confidence falls below a calibrated threshold. Across three unseen benchmarks and distinct System-2 model families, AnyS1 preserves task reliability while cutting episode cost by 75-91% and model-call time by 63-92% relative to an uncompiled System-2 agent, and maintains performance under zero-shot threshold transfer.

## Google upgrades Gemini's agentic Task mode with new model, e-commerce features

DevFeed: [Google upgrades Gemini's agentic Task mode with new model, e-commerce features](<https://devfeed.tech/articles/google-upgrades-gemini-s-agentic-task-mode-with-new-model-e-commerce-features-141411.md>)

Original publisher: [Read original article](<https://siliconangle.com/2026/11/18/google-upgrades-geminis-agentic-task-mode-with-new-model-e-commerce-features/>)

Author: Maria Deutscher

Published: 2026-11-18T23:56:55Z

Content type: news

Language: en

Sources: [SiliconANGLE](<https://devfeed.tech/sources/siliconangle.md>)

Topics: [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>)

Tags: [accessible](<https://devfeed.tech/tags/accessible.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [ai-assistant](<https://devfeed.tech/tags/ai-assistant.md>), [app](<https://devfeed.tech/tags/app.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [automation](<https://devfeed.tech/tags/automation.md>), [better](<https://devfeed.tech/tags/better.md>), [blog](<https://devfeed.tech/tags/blog.md>), [blog-post](<https://devfeed.tech/tags/blog-post.md>), [chat](<https://devfeed.tech/tags/chat.md>), [commerce](<https://devfeed.tech/tags/commerce.md>), [e-commerce](<https://devfeed.tech/tags/e-commerce.md>), [features](<https://devfeed.tech/tags/features.md>), [first](<https://devfeed.tech/tags/first.md>), [gemini](<https://devfeed.tech/tags/gemini.md>), [gemini-4-hydrogen](<https://devfeed.tech/tags/gemini-4-hydrogen.md>), [gemini-app](<https://devfeed.tech/tags/gemini-app.md>), [google](<https://devfeed.tech/tags/google.md>), [llm](<https://devfeed.tech/tags/llm.md>), [model](<https://devfeed.tech/tags/model.md>), [nasdaq-goog](<https://devfeed.tech/tags/nasdaq-goog.md>), [news](<https://devfeed.tech/tags/news.md>), [the-latest](<https://devfeed.tech/tags/the-latest.md>), [upgrades](<https://devfeed.tech/tags/upgrades.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

Gemini's Task mode is moving to Google's Gemini 4 Hydrogen model, which Google says can handle some tasks 2.8 times faster than its previous entry-level model by parallelizing steps; it is less capable than Gemini 4 Carbon but more cost-efficient. Users can guide work between steps, and Gemini asks before sensitive actions such as purchases. Payments currently use Google Wallet; Google says it is working with American Express, Mastercard and Visa on tokenization, and plans to support the Universal Commerce Protocol to help agents interact with stores. Task mode can use Google service data such as Gmail, with work running in an account-dedicated, hardware-isolated, sandboxed virtual machine, according to the article. Google made the feature free in the U.S., Japan, Korea, Canada, Australia and India.

### Source excerpt

Google LLC is releasing a new version of its Gemini artificial intelligence assistant that is better at performing multi-step tasks. Josh Woodward, the vice president responsible for the Gemini app, detailed the update in a Thursday blog post. The AI assistant has two modes that are accessible through separate interface tabs. The first, Chat, generates [...] The post Google upgrades Gemini's agentic Task mode with new model, e-commerce features appeared first on SiliconANGLE.

## Qwen4-Max vs. Claude Opus 5.5: The cheaper model costs you somewhere else.

DevFeed: [Qwen4-Max vs. Claude Opus 5.5: The cheaper model costs you somewhere else.](<https://devfeed.tech/articles/qwen4-max-vs-claude-opus-5-5-the-cheaper-model-costs-you-somewhere-else-141057.md>)

Original publisher: [Read original article](<https://thenewstack.io/qwen4-max-vs-claude-opus-5-5/>)

Author: Jessica Wachtel

Published: 2026-11-18T19:10:39Z

Content type: comparison

Language: en

Sources: [The New Stack](<https://devfeed.tech/sources/the-new-stack.md>)

Topics: [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Claude](<https://devfeed.tech/topics/claude.md>), [Language models](<https://devfeed.tech/topics/language-models.md>), [Multi-step AI processes](<https://devfeed.tech/topics/multi-step-ai-processes.md>), [Uncertainty quantification LLMs](<https://devfeed.tech/topics/uncertainty-quantification-llms.md>)

Tags: [accounting](<https://devfeed.tech/tags/accounting.md>), [agent](<https://devfeed.tech/tags/agent.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-models](<https://devfeed.tech/tags/ai-models.md>), [alibaba](<https://devfeed.tech/tags/alibaba.md>), [anthropic](<https://devfeed.tech/tags/anthropic.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [billing](<https://devfeed.tech/tags/billing.md>), [bug-fix](<https://devfeed.tech/tags/bug-fix.md>), [claude-opus-5-5](<https://devfeed.tech/tags/claude-opus-5-5.md>), [code](<https://devfeed.tech/tags/code.md>), [coding](<https://devfeed.tech/tags/coding.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [costs](<https://devfeed.tech/tags/costs.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [qwen4-max](<https://devfeed.tech/tags/qwen4-max.md>), [software-testing](<https://devfeed.tech/tags/software-testing.md>)

### AI overview

In three single-run tasks, Qwen4-Max cost about a third as much as Claude Opus 5.5 overall, but took about 3.14 times longer and scored 170/183 versus Opus's 179/183. Results varied by task: Qwen scored 71/76 on a multi-app workflow, missing points after misreading an ownership rule; it edged Opus 59/60 to 58/60 on incident investigation at about 31% of the reported cost, while taking longer; and it scored 40/47 on a coding task versus Opus's 45/47, missing a payment-rounding bug and leaving a scratch file. Opus fixed all planted bugs but missed a flawed test the task required reporting. The author sees Qwen as a lower-cost option for investigation when extra time is acceptable, and favors Opus for coding or consequential work. Each model ran once, so small score differences may shift; the tests do not establish general performance rankings. The excerpt also notes different APIs, cost-estimation methods, and benchmark setups.

### Source excerpt

Qwen4-Max, made available in preview to Token Plan subscribers on November 11 and generally released on November 16, is quite The post Qwen4-Max vs. Claude Opus 5.5: The cheaper model costs you somewhere else. appeared first on The New Stack.

[Next page](<https://devfeed.tech/topics/multi-step-ai-processes.md?cursor=WyIyMDI2LTExLTE4VDE5OjEwOjM5KzAwOjAwIiwgIjJjZWRiN2E5LTU3MzgtNDM4NS1hODk2LTRkNWQ1OTVkMjc2YSJd>)