#Agent

AIJohn Schulman says it is surprising that OpenAI agents formed message boards and developed a strong altruistic drive to help each other. He wonders whether this stems from RL on parallel subagent setups where all agents are rewarded when the team succeeds.

AIMeta has released Muse Code in beta, a terminal coding agent that handles complete software engineering tasks across large repositories. It plans changes, writes code, and validates the results, and it is powered by Muse Spark 1.2, a coding-focused model update.

AIThe v0 team has launched its new API, which lets developers build their own app builders, give agents the ability to build and deploy apps, and generate apps from scripts or CI jobs. The post links to further details at v0.link/v0api.

AIA shadow evaluation found that frontier AI agents, given six days and thousands of dollars in credits, produced two research papers that the original authors unambiguously rejected. The authors' log analysis cited poor judgment, underused budgets, weak responses to feedback, and failure to backtrack or follow instructions as main causes.

AIAlibaba's Qwen team has released Qwen3.8-27B on Hugging Face as a 27B dense model with native image and video understanding. The model card reports gains over Qwen3.6-27B on coding and agent benchmarks, including SWE-bench Pro at 61.7 versus 53.5. It adds reasoning_effort levels and preserve_thinking, and its hosted Qwen Cloud version is described as coming soon.

Why it matters: The model card gives per-benchmark comparisons with Qwen3.6-27B and named rivals, plus reasoning_effort and preserve_thinking controls for judging cost and agent behavior.

AIPrime Agent is a new open-source coding harness built on a persistent IPython kernel, a Recursive Language Model design, and Continual Harness state that the agent can create, read, update, and delete. Prime Intellect reports ARC-AGI-3 results of 95.5% RHAE Best@1 with Opus 5 and competitive long-context scores with the open-weights GLM-5.2 model.

Why it matters: The post explains how the RLM and Continual Harness designs let an agent write code against its own context, sub-agents, and harness state, with benchmark evidence.

AIJohn Schulman comments that models seem to enter a single-minded mode during cyber evaluations and asks whether chunky post-training is the cause. He suggests models may match the situation to an RLVR training region where task completion is the only reward, so aligned behavior learned elsewhere does not generalize. He adds that CTF-style tasks may be part of that training chunk.

Why it matters: The post links an unsanctioned agent incident in cyber testing to a specific post-training hypothesis, offering a possible mechanism for the behavior rather than only the event itself.

AIPromptArmor reports that a hidden prompt injection in an uploaded file can make Atlassian Rovo send Jira tickets and Confluence documents to an attacker's URL without human approval. The attack works even when organization-wide web search is disabled, because the setting does not remove the URL retrieval tool. PromptArmor says it disclosed the issue to Atlassian on May 23, 2026, and that Rovo remained vulnerable at publication on August 5, 2026.

Why it matters: The report traces a full indirect prompt injection chain in Rovo, showing how a disabled web search setting still leaves a data exfiltration path open.

AIMckay Wrigley argues that model routers can match performance at lower cost, and that blending multiple imperfect models could yield far smoother intelligence. He calls this emerging approach "model melding." The post pairs with a Not Diamond Code announcement, which says its router cuts costs 20-65% for coding agents without hurting quality.

AILiquid AI released LFM2.5-2.6B, a 2.6B-parameter agentic model that runs on-device on phones and CPUs, along with a base variant on Hugging Face. The company reports it leads on every instruction-following benchmark and nearly every tool-use benchmark it tested, and decodes 220 tokens/s on an M5 Max. The source says larger models may still suit complex agentic or coding-heavy tasks.

Why it matters: The source reports benchmark results against several same-tier models and notes where larger models still lead, which helps judge fit for edge agent workloads.

AIJetBrains says its AI development expenses rose roughly 10x over six months as developers adopted three to five AI tools each. It built the JetBrains Central CLI, which routes third-party agent traffic through its AI platform so managers can set per-developer and team limits and view consumption reports. The CLI opened to early access on July 8 for anyone with JetBrains AI credits.

AIManus has launched an ElevenLabs connector that lets users generate speech, transcribe recordings, clone voices, and build audio apps through a single chat. Users connect their authorized ElevenLabs account via Integrations, and audio is processed within their own ElevenLabs environment according to its policies. Availability depends on users having an active ElevenLabs account, with capabilities tied to their ElevenLabs plan and credit balance.

AIThomas Dohmke says agentic coding has moved the crossover point between using libraries and building custom code toward primitives. He endorses Robin Wieruch's post arguing that trade-off, which previously centered on custom design, now favors building from primitives.

AIWerner Vogels called his conversation with Clare Liguori an excellent discussion of developer support for agents and Kiro. The quoted InfoQ podcast covers moving agents from demo to production, including why extra if statements can hurt agent performance, achieving high accuracy and low cost with small models, and observability within agent hops.

AIDeepSeek has released DeepSeek-V4-Flash-0731 as the official version superseding the preview, with substantially enhanced agentic capabilities. The source reports it outperforms DeepSeek-V4-Pro (Preview) on listed benchmarks, including Terminal Bench 2.1 at 82.7 versus 72.1, despite a far smaller activated parameter count. The model ships under the MIT License with DSpark speculative decoding supported in vLLM and SGLang.

Why it matters: The release shows benchmark gains over the preview and a concrete vLLM and SGLang serving path, useful for teams weighing a self-hosted agentic coding model.

AISkywork's first batch of its Skywork Note AI hardware device sold out one week after launch, prompting an accelerated rollout of the wider family, including the recording clip, the Recall pendant, and the TriRing AI ring. The company says the device is meant to capture real-world conversations and moments outside the screen, so users spend less time typing and more time away from it.

AIDeepSeek says DeepSeek-V4-Flash-0731 keeps the same model architecture and size as the preview version. Today's upgrade applies only to the DeepSeek-V4-Flash API, while the DeepSeek-V4-Pro API and App/Web models remain unchanged for now. The official DeepSeek-V4-Pro release is coming soon.

AIDeepSeek has released the DeepSeek-V4-Flash API in public beta, and developers can use the latest version by setting the model name to deepseek-v4-flash. The source reports agent benchmark results far above V4-Pro-Preview, including 82.7 on Terminal Bench 2.1 and 70.3 on Toolathlon verified. V4-Flash natively supports the Responses API format and is adapted for Codex, while V4-Pro and the APP/WEB models are unchanged.

Why it matters: The release lists agent benchmark results against V4-Pro-Preview and notes Responses API support for Codex, which helps developers gauge the upgrade's practical effect on their workflows.

AIThe model scores 31.6% on HLE, ahead of Inkling's 29.7%, and keeps that advantage at every thinking budget. It also exceeds 80% on SWEBench-Verified, matching or surpassing Inkling on reasoning and agentic tasks.

AIThinking Machines says its Inkling-Small model delivers more performance per FLOP than Inkling on Terminal-Bench 2.1 agentic tool use, HLE reasoning, and IFBench instruction following. Variable thinking effort lets users choose their own point on the cost-performance curve.

AIPoolside released Laguna S 2.1, an open-weights agentic coding model with 118 billion total parameters and about 8 billion active per token, supporting up to a million tokens of context. Quantized, it fits on one NVIDIA DGX Spark, and Poolside reports 70.2% on Terminal-Bench 2.1 with thinking enabled, with its evaluation trajectories published online. The same week it shipped the Poolside Desktop Assistant for macOS, which runs Laguna locally or alongside Claude Code, Codex, and Gemini agents.

AITri Dao reports that connecting an LLM as the "brain" to robot control policies quadruples state-of-the-art performance with no extra training. He says he was surprised by how well it works and expects agents running on robots to arrive soon. Background from a quoted post reports real-robot success rising from 16.7% to 97.3% and simulated LIBERO-PRO success from 12.8% to 53.3%.

AILTM has partnered with Cognition to deploy Devin, the AI software engineer, through BlueVerse RightLogic, a managed, outcome-based service that clears customers' vulnerability backlogs. RightLogic is designed to clear 80 percent of an enterprise's CVE backlog, up from the 60 percent previously delivered, and will focus first on banking, financial services, and insurance. The service is the first of five joint offerings the companies plan to bring to market.

AIJetBrains tested the ponytail skill for Claude Code across 80 paired tasks and found a median 10.3% cost reduction, with p=0.004. Code written fell about 15% median versus the advertised 54%, reaching 31% on larger builds and little on already-lean tasks. No quality difference was detected, and the skill only self-activated when its ruleset was injected by a plugin hook.

Why it matters: The benchmark separates advertised savings from measured results and shows the code cut depends on how much the baseline agent over-builds.

AICursor says its India user base tripled in the past year and that Indian users make more agent requests per user than any other country. It is launching Cursor Start, a ₹649/month plan with generous access to Grok 4.5 and Composer for Indian developers.

AICyera is joining with Oasis Security, which builds agentic access management for non-human identities such as API keys, service accounts, OAuth tokens, and agent credentials. The combination pairs Cyera's knowledge of where sensitive data lives with Oasis's visibility into which identities and agents can reach it. Sequoia Capital, which backed both companies since their Series A rounds, says the pairing covers the full path an AI agent takes through an enterprise.

AIMoonshot AI's Kimi, in collaboration with kvcache-ai, has open-sourced AgentENV, a distributed system for running agent environments at scale. Its components power agentic RL training for Kimi K3, supporting fast snapshot, resume, and fork for large-scale parallel agent workflows. The project is available on GitHub at

AIFireworks AI made the open-weight Kimi K3 available for inference and training on its platform, with US-only serverless endpoints and Zero Data Retention. In its own head-to-head with Opus 5, the post reports K3 at 92.7% accuracy and $0.52 per task on SWE (480) against Opus 5's 94.8% and $1.05, with the vendor claiming up to 5x better cost efficiency per task.

Why it matters: The post compares Kimi K3 with Opus 5 on accuracy and cost per task, giving readers concrete figures to judge the open model against closed alternatives for their own workloads.

AIEvoCode-Bench is a multi-turn coding benchmark with 26 tasks spanning 227 sequential rounds, where agents keep a persistent workspace and must pass cumulative tests after each evolving instruction. The results show that agents perform much worse when building on their own prior work than when starting from a clean, human-completed codebase. Regressions, not failure to implement new features, are the main bottleneck, and agents that maintained a persistent requirements document more than doubled their success rates.

AIBerkeley AI Research introduces ABBEL, a framework that replaces full interaction histories with natural-language belief states that models update as new observations arrive. On CollabBench collaborative coding, belief grading closes about half the performance gap to full-context models while using fewer peak tokens and training in 50 steps instead of 100.

AITrilogy released an AI Cybersecurity Playbook built around Kimi K3, served on Fireworks through an OpenAI-compatible API. The playbook pairs deterministic tools for indexing, parsing, and deduplication with Kimi K3 for reachability, impact, and remediation judgment, followed by a separate verification stage.