AI Agents

OpenAI has added its new ChatGPT Voice experience to the ChatGPT desktop app, enabling users to direct ChatGPT Work, Codex, and computer-use capabilities by speaking. The update shifts voice from conversational input toward a way to coordinate multi-step AI tasks.

AgentDebugX is an open-source framework for observing, attributing, and repairing failures in LLM agents. Its central idea is that the step where an error appears is often not the step that caused it.

DataFlow-Harness addresses the gap between natural-language requests and persistent data-pipeline artifacts. Instead of asking an LLM agent to write loose scripts, it guides the agent to construct platform-native DAGs through typed, incremental updates.

NexForge tackles a core bottleneck in agent post-training: scalable executable data. Instead of building tasks around fixed tools or repositories, it starts from high-level capability requirements and generates runnable tasks plus expert trajectories.

Rather than scaling a general model, this paper builds a training ground where search agents can check their own progress and recover from mistakes. The core idea is self-improvement through verifiable interaction traces.

API-calling agents usually need lots of high-quality trajectories, but building full executable environments is expensive and slow. This paper shows that LLMs can simulate stateful API interactions from specs alone.

A new paper argues that encrypted negotiation values are not enough: concession timing, offer trajectories, and convergence patterns can still reveal private constraints. The authors propose an adaptive randomized policy to reduce inference attacks while preserving negotiation performance.

RESOURCE2SKILL proposes a way to distill videos, repositories, articles, and reference artifacts into reusable skills that software agents can retrieve and execute.

Cura 1T treats healthcare as a connected set of skills, not a single medical QA problem. Its training loop aims to improve consultation, reasoning, diagnosis, and tool use together.

Recursive Harness Self-Improvement reframes the agent harness as something that can be optimized, not merely hand-written once. The paper reports that a few lightweight iterations can improve low-reasoning-effort agents while cutting inference cost.

Kimi K3 arrived quickly after launch and was added to the model marketplace on the same day. What stands out most is not only its size, but its breadth across coding, long-horizon tasks, search, tables, and front-end work.

A VentureBeat Pulse Research survey of 107 enterprises suggests that AI agents are gaining real access to systems and data faster than security controls can catch up. More than half of respondents have already seen a confirmed incident or a near miss.

1Password has launched a Claude browser integration that lets users authorize the AI agent to use saved login credentials. The company says Claude can complete sign-in steps without seeing passwords or MFA one-time codes.

A new arXiv survey frames self-improving agents as adaptive systems that turn experience into accumulated capability gains. Its focus is not a single algorithm, but the full configuration of models, prompts, memory, tools, and control logic.

A new arXiv paper proposes an AI-native insurance framework for agentic AI deployments, linking autonomy, permissions, governance, and dependency risk to underwriting and contract design. The work frames insurance as both an operational cost and a potential governance mechanism for autonomous AI systems.

Ai2’s Shippy shows that production agents in high-stakes domains are not just model wrappers. Reliability comes from tool design, sandboxing, explicit limits, and workflow-level evaluation.

A new arXiv paper argues that one-shot benchmark gains are not enough to judge agent optimizers. In a two-phase continual-learning setup, methods diverged sharply once new tasks were introduced.

A new arXiv paper proposes the DVM-HALL framework and the Net Human-Agent Score to rethink loyalty in autonomous commerce. The model treats trust, delegation, machine experience, and verifiable execution risk as core parts of brand choice.

A new arXiv paper introduces Experience Memory Graph, a framework that stores failure fixes as structured graph edits. Instead of repeated reflection loops, agents retrieve relevant correction patterns and act once.

SPyCE proposes a training framework where a multimodal agent’s policy and skill library improve together. Instead of treating trajectories only as rewards or static memories, it distills them into reusable execution and workflow skills.

A new arXiv paper surveys 21 proposed permission systems for AI agents and compares them with five commercial agents. Its central argument: agent safety needs user-level permissions, not just one-size-fits-all product policies.