Search “Agents”

Cognition says Devin Fusion scored 68 on the Coding Agent Index

On October 20, 2026, Cognition announced that Devin Fusion scored 68 on the ArtificialAnlys Coding Agent Index. According to Cognition, the result was led by Opus 5.5 with assistance from GPT-6 Luna. Cognition also claims that Devin Fusion outperforms Claude Code paired with Opus 5.5 on both performance and price, with a 31% lower cost per task. The above score and cost comparison are Cognition's unilateral claims; the input materials contain no independent third-party verification or subsequent developments.

Cognition says Devin Fusion scored 68 on the Coding Agent Index

On October 20, 2026, Cognition announced that Devin Fusion scored 68 on the ArtificialAnlys Coding Agent Index. According to Cognition, the result was led by Opus 5.5 with assistance from GPT-6 Luna. Cognition also claims that Devin Fusion outperforms Claude Code paired with Opus 5.5 on both performance and price, with a 31% lower cost per task. The above score and cost comparison are Cognition's unilateral claims; the input materials contain no independent third-party verification or subsequent developments.

View story:Cognition says Devin Fusion scored 68 on the Coding Agent Index

David Williams calls for attention to the immediate risks of Agentic AI

On October 19, 2026, FIOR Group CEO David Williams proposed that enterprises should focus on the immediate, real-world risks brought by Agentic AI, rather than being swayed by exaggerated AI doomsday narratives. Citing a Cloud Security Alliance survey, he said that 82% of organizations discovered AI agents operating in the past year without the knowledge of their security or IT teams, and only 11% of organizations automatically block an agent's operations when it exceeds its authority. He advocates establishing a deterministic execution layer outside the model, assigning an accountable owner and permission scope to each agent, and completing verification and auditing before actions are executed. This view is currently reported by Unite.AI, and there are no subsequent developments of fact yet.

Lauren proposes two cognitive divides regarding Agents

On October 17, 2026, lauren asked: Is chat the ultimate form of AI agent UI? and invited everyone to share their reasons. On October 19, lauren launched a poll on the same question, and about 70% of respondents answered "no"; she believes the disagreement lies in two imaginings: one sees agents as an extension of one's own mind, and in the future brain-computer interfaces may eliminate the need for an input box; the other sees agents as independent beings capable of making decisions autonomously and working alongside us. She also believes that voice and text chat are essentially the same, both treating the agent as an "other" to converse with. On the same day, swyx responded in the dialogue style of the Bhagavad Gita, saying that people on X always think they are the first to want to end chat, and that people say they want to end chat but in fact return to it in even greater numbers; what they truly desire is not truth but "newness." The discussion currently revolves around this question.

View story:lauren asks whether chat is the ultimate form of Agent UI

Vercel AI Gateway provides a unified Agent session view

On October 19, 2026, Vercel AI Gateway provides a unified session view across personal and coding agents, apps, and API keys. According to the Vercel developer account, AI Gateway logs can now show each agent session, group requests by conversation, and display cost, cache hit rate, and the agent that executed the session; session affinity pins each session to the provider where its prompt cache resides. Guillermo Rauch reposted and praised this unified view, saying it covers personal and coding agents, apps, and API keys, and emphasized that Vercel AI Gateway gives users control over model choice, data (ZDR), and the metadata used to understand RoI. There are currently no follow-up developments.

The author announced that the practical experience had been compiled into "Eight Lectures on Agent Engineering in Practice"

On October 18, 2026, the author announced that the practical experience had been compiled into "Eight Lectures on Agent Engineering in Practice". On the same day, the Tencent Cloud Developer Community published an introduction to the series, describing how it focuses on upgrading prototypes that can call models and tools into systems that can complete business tasks, explain failures, control costs, and iterate continuously. The introduction states that the entire series runs through two fictional teaching cases—a case-filing assistant Agent and a postoperative follow-up Agent—covering eight topics: task contracts, reasoning and collaboration, context and memory, reliable tool execution, observability and replay, evaluation and calibration, security boundaries, and production operations; the text emphasizes that a model being able to suggest the next step does not mean the system can deliver business results, and that the example fields, budgets, and figures are all teaching calculations and do not constitute legal advice. Currently, only the introduction to the series has been published, and no factual evidence has yet been seen for the content of the subsequent lectures.

Arena announces DeepSeek-V4.1-Flash-1017 tops the Agent Arena open-source leaderboard

On October 17, 2026, DeepSeek released DeepSeek-V4.1-Flash-1017, which the company says delivers stronger benchmark performance, faster task completion, and shorter outputs, achieving a Pareto improvement in performance and cost; it is now available on the official website, App, and API, and the model also launched on OpenRouter the same day. The company also announced that the Search and Fetch APIs are open to everyone, priced at $0.75 per 1,000 calls per endpoint, with calls made through server-side tools in DeepSeek Harness, Claude Code, and Codex temporarily free, and billing starting at 16:00 UTC on October 31, 2026. On October 17, Thomas Wolf commented that DeepSeek called this a "minor update," yet Terminal-Bench 4.0 rose from 31.2 to 42.3. On October 19, Arena announced that the model, with a score of 1670, became the highest-ranked open-source model on Code Arena: WebDev. On October 20, OpenCode updated it to version 1017 in Go and Free; Unsloth released a GGUF that can run locally, with the 1-bit quantized version retaining about 88.2% accuracy and shrinking in size from 508GB to 250GB; Arena then announced that it topped the Agent Arena open-source model leaderboard.

View story:DeepSeek releases DeepSeek-V4.1-Flash-1017

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.

David Williams argues for focusing on the immediate risks of Agentic AI

On October 19, 2026, FIOR Group CEO David Williams proposed that enterprises should focus on the immediate, real-world risks brought by Agentic AI, rather than being swayed by exaggerated AI doomsday narratives. Citing a Cloud Security Alliance survey, he said that 82% of organizations discovered AI agents operating in the past year without the knowledge of their security or IT teams, and only 11% of organizations automatically block an agent's operations when it exceeds its authority. He advocates establishing a deterministic execution layer outside the model, assigning an accountable owner and permission scope to each agent, and completing verification and auditing before actions are executed. This view is currently reported by Unite.AI, and there are no subsequent developments of fact yet.

View story:David Williams calls for attention to the immediate risks of Agentic AI

English translation is being prepared.

View story:English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.

Legora uses Modal Sandboxes to run 2,000 agent evaluations in parallel

On October 19, 2026, the official Modal engineering blog published a post introducing that Legora places each evaluation run in an independent Modal Sandbox; a single experiment can run more than 2,000 times in parallel, used to test legal agents that span thousands of documents and run for several hours. The team uses Modal Volumes to snapshot documents, databases, and permission states, and by building images hourly and pre-warming Sandboxes, shortens reset-to-commit from a full build to an incremental update. After migrating to Modal VM Sandboxes, Docker works inside the sandbox like an ordinary Linux host, and Legora removed all of its previous network and FUSE workarounds.

View story:Legora uses Modal Sandboxes to run 2,000 agent evaluations in parallel

Vercel AI Gateway provides a unified view of agent sessions

On October 19, 2026, Vercel AI Gateway provides a unified session view across personal and coding agents, apps, and API keys. According to the Vercel developer account, AI Gateway logs can now show each agent session, group requests by conversation, and display cost, cache hit rate, and the agent that executed the session; session affinity pins each session to the provider where its prompt cache resides. Guillermo Rauch reposted and praised this unified view, saying it covers personal and coding agents, apps, and API keys, and emphasized that Vercel AI Gateway gives users control over model choice, data (ZDR), and the metadata used to understand RoI. There are currently no follow-up developments.

View story:Vercel AI Gateway provides a unified Agent session view

Legora uses Modal Sandboxes to run 2,000 agent evaluations in parallel

On October 19, 2026, the official Modal engineering blog published a post introducing that Legora places each evaluation run in an independent Modal Sandbox; a single experiment can run more than 2,000 times in parallel, used to test legal agents that span thousands of documents and run for several hours. The team uses Modal Volumes to snapshot documents, databases, and permission states, and by building images hourly and pre-warming Sandboxes, shortens reset-to-commit from a full build to an incremental update. After migrating to Modal VM Sandboxes, Docker works inside the sandbox like an ordinary Linux host, and Legora removed all of its previous network and FUSE workarounds.

The author announces that practical experience has been compiled into "Eight Lessons on Agent Engineering in Practice"

On October 18, 2026, the author announced that the practical experience had been compiled into "Eight Lectures on Agent Engineering in Practice". On the same day, the Tencent Cloud Developer Community published an introduction to the series, describing how it focuses on upgrading prototypes that can call models and tools into systems that can complete business tasks, explain failures, control costs, and iterate continuously. The introduction states that the entire series runs through two fictional teaching cases—a case-filing assistant Agent and a postoperative follow-up Agent—covering eight topics: task contracts, reasoning and collaboration, context and memory, reliable tool execution, observability and replay, evaluation and calibration, security boundaries, and production operations; the text emphasizes that a model being able to suggest the next step does not mean the system can deliver business results, and that the example fields, budgets, and figures are all teaching calculations and do not constitute legal advice. Currently, only the introduction to the series has been published, and no factual evidence has yet been seen for the content of the subsequent lectures.

View story:The author announced that the practical experience had been compiled into "Eight Lectures on Agent Engineering in Practice"

English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.