Search “Agents”

Aaron Levie predicts that Agent token consumption will reach 1,000 times the current level

Box CEO Aaron Levie predicted on October 29, 2026, that agents will in the future consume 1,000 times the tokens they do today, because agents will need to run within workflows, process enterprise data, build almost all software, and secure enterprises. Levie believes that achieving this scale requires simultaneously advancing the capabilities, safety, and efficiency under large workloads of both closed-source models and open-weight models. He also mentioned that Suhail is building Particle for this purpose; Suhail announced the new company Particle the same day, whose particle-1 model uses about 45% fewer tokens at comparable performance.

Aaron Levie predicts that Agents will consume a thousand times more tokens than today

Box CEO Aaron Levie predicted on October 29, 2026, that agents will in the future consume 1,000 times the tokens they do today, because agents will need to run within workflows, process enterprise data, build almost all software, and secure enterprises. Levie believes that achieving this scale requires simultaneously advancing the capabilities, safety, and efficiency under large workloads of both closed-source models and open-weight models. He also mentioned that Suhail is building Particle for this purpose; Suhail announced the new company Particle the same day, whose particle-1 model uses about 45% fewer tokens at comparable performance.

View story:Aaron Levie predicts that Agent token consumption will reach 1,000 times the current level

Thariq proposes that technical and non-technical people have opposite mindsets toward Agents

Thariq offered an observation: technical and non-technical people exhibit opposite mindsets when facing AI agents. Technical people feel frustrated, believing their skills have become useless; non-technical people are eager to understand technology better so they can use AI agents more effectively. This observation was published on October 29, 2026, and there are currently no subsequent developments.

Thariq proposes that technical and non-technical people have opposite mindsets toward Agents

Thariq offered an observation: technical and non-technical people exhibit opposite mindsets when facing AI agents. Technical people feel frustrated, believing their skills have become useless; non-technical people are eager to understand technology better so they can use AI agents more effectively. This observation was published on October 29, 2026, and there are currently no subsequent developments.

View story:Thariq proposes that technical and non-technical people have opposite mindsets toward Agents

MiniMax Code team open-sources Agent workflow and Skills

On October 28, 2026, the MiniMax Code team open-sourced MCode Stack, packaging the Agent workflows and Skills the team uses daily into installable SKILL.md files and manifests, which can be integrated as plugins into MiniMax Code and any Agents that support Skills. The repository packages stacks by member, among which stacks/ronny/ is the most complete, containing 15 skills covering sub-Agent decomposition, context and tool budget auditing, long-session handoff, independent review, and hotfix judgment. An accompanying notes/ column, MCode Engineering Notes, explains the rationale; six engineering-* entries are adapted from Google Engineering Practices, with the adapted portions retaining CC-BY 3.0 and the rest under MIT. The next day, reports stated that the project is open-sourced under the MIT license, includes 17 SKILL.md skills and 5 manifests, does not contain MiniMax Code source code, and can be installed by copying the skill directories, via a local plugin, or through a plugin marketplace, and can also be installed in Claude Code; the original text did not publish evaluation results.

Arena.ai says Gemini 4 Carbon ranks second on Agent Arena

On October 28, 2026, Google DeepMind released Gemini 4 Carbon, calling it its strongest model, ranking first with 59 points on the Artificial Analysis Intelligence Index v5.0, and scoring 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench. The model is aimed at long-horizon engineering tasks and was first opened to some cybersecurity defenders before being expanded through the Fairwind Program. On October 29, Arena.ai said Gemini 4 Carbon ranked 2nd on Agent Arena with a net improvement of 16.2% and pushed the Pareto frontier forward; the same day, Yuchen Jin cited Artificial Analysis data to raise reservations, noting that the evaluation used the Antigravity CLI, and he planned to try it instead in Codex and Omnigent. On October 30, the Artificial Analysis Intelligence Index showed Carbon in first place with 59 points, 1 point higher than Claude Opus 5.5. On October 31, Perplexity CEO Aravind Srinivas shared an early evaluation, saying it set a new SoTA on the WANDR benchmark. There is also news that

View story:Google releases the Gemini 4 Carbon model

Thariq proposed hoping to express ideas to the Agent using sketches

On October 29, 2026, Thariq proposed that he very much hopes to first sketch out animation ideas and character concept drafts himself, and then have the agent refine them, rather than going to the trouble of searching for reference cases. He also mentioned that he has seen artists deeply worried about AI, but he still hopes to use the agent to complete his ideas, and lamented his own lack of this expressive ability. Currently, this statement exists only in one remark by Thariq himself, with no subsequent developments or institutional response.

Snaptrude launches the AI agent Snaptrude Agent

Snaptrude launches the AI agent Snaptrude Agent, which can turn a design brief into a construction-grade 3D model covering structure, MEP and drawings. According to Testing Catalog, the agent first presents a plan before modeling and producing drawings, and everything it adds remains editable, so architects can take over at any time. Rohan Paul relays that the model can start from a site or a brief to generate an LOD 400 construction-grade model including structure, MEP and complete drawings, and supports exporting back to the original software while preserving component families. According to its founder, during the early access phase more than 2,000 designers have run over 25,000 tasks through the agent, and the team has also released AECBench, a benchmark for AEC, along with its first results.

Snaptrude launches AI agent Snaptrude Agent

Snaptrude launches the AI agent Snaptrude Agent, which can turn a design brief into a construction-grade 3D model covering structure, MEP and drawings. According to Testing Catalog, the agent first presents a plan before modeling and producing drawings, and everything it adds remains editable, so architects can take over at any time. Rohan Paul relays that the model can start from a site or a brief to generate an LOD 400 construction-grade model including structure, MEP and complete drawings, and supports exporting back to the original software while preserving component families. According to its founder, during the early access phase more than 2,000 designers have run over 25,000 tasks through the agent, and the team has also released AECBench, a benchmark for AEC, along with its first results.

View story:Snaptrude launches the AI agent Snaptrude Agent

OpenRouter announced the return of Royale: Last Agent Standing

OpenRouter announced the return of Royale: Last Agent Standing. According to the information it released, 20 models from 10 labs will compete in 30 matches in a livestream, with only one winner left in the end; the livestream commentary will be provided by ElevenLabs V4, on Thursday, October 29 at 12 PM ET. The same report noted that the cited teaser post stated the scale as 30 models and 20 labs, which is inconsistent with the scale described in the main text. There are currently no subsequent developments.

AgentWeb goes live: AI agents complete website tasks without a browser

On October 29, 2026, AgentWeb released web infrastructure for intelligent agents, enabling AI agents to complete website tasks without a browser by invoking preset actions instead of reading screenshots and clicking buttons. According to a report by Rohan Paul the same day, the company claims it has completed 47/47 real-time tasks across 15 government portals, 7 insurance companies, and 25 sites requiring login such as Amazon and LinkedIn, with a median time of 1.9 seconds and a total cost of 3 US dollars per round. Each mapped task is exposed as a tool, including an input schema, preconditions, permission boundaries, and completion credentials, and can be invoked from any assistant. There are currently no subsequent developments.

indigo previews that the end-of-month livestream will discuss evolving toward the Agentic Internet

On October 29, 2026, indigo published a post previewing the theme of this month's end-of-month livestream: "Evolving Toward the Agentic Internet." indigo said they had originally planned to keep it simple, but the content became super rich, and invited readers to check the table of contents and reserve a spot; the livestream will go live on YouTube over the weekend. There is currently no information on further developments.

Tencent, Tencent Cloud and WorkBuddy jointly release an AI Agent enterprise solutions handbook

On October 29, 2026, Tencent, Tencent Cloud and WorkBuddy jointly released the 78-page "2026 AI Agent Enterprise Solutions and Implementation Practice Handbook." According to a report by Shao Meng that day, the handbook uses WorkBuddy as the entry point, advancing personal assistants into "digital employees" that enter business systems, with ClawPro, Agent Runtime and TokenHub responsible for provisioning, isolation, models and governance. The handbook proposes a three-stage framework of "able to converse → able to work → able to take up a post," and presents case data including Yuzu's Agent penetration rate exceeding 75%, Youka's MTTR of 5–15 minutes, and operations and maintenance costs of approximately -40%, all of which are self-reported in the text. There are currently no subsequent developments.

meng shao says Video Agent products may be reduced to token relay stations

On October 29, 2026, Shao Meng put forward a judgment: although domestic Video Agent teams such as LibTV, TapNow, Flova, Pexo, and Medeo emphasize barriers such as ecosystem, Agent architecture, and industry know-how in financing roadshows, what they end up building may all turn out to be token relay stations. He speculated that if these websites supported BYOK, the cost for users to bring their own video generation model API Key would be lower than consuming Credits on the site, and even lower when the API Key has a discount. He suggested using REA or building one's own Video Agent from scratch paired with discounted API Keys to reduce costs, believing that for users making video content on TikTok, Ins, and Douyin, the money saved would be considerable. Currently, this judgment has no subsequent developments.

MiniMax Code team open-sources Agent workflows and Skills

On October 28, 2026, the MiniMax Code team open-sourced MCode Stack, packaging the Agent workflows and Skills the team uses daily into installable SKILL.md files and manifests, which can be integrated as plugins into MiniMax Code and any Agents that support Skills. The repository packages stacks by member, among which stacks/ronny/ is the most complete, containing 15 skills covering sub-Agent decomposition, context and tool budget auditing, long-session handoff, independent review, and hotfix judgment. An accompanying notes/ column, MCode Engineering Notes, explains the rationale; six engineering-* entries are adapted from Google Engineering Practices, with the adapted portions retaining CC-BY 3.0 and the rest under MIT. The next day, reports stated that the project is open-sourced under the MIT license, includes 17 SKILL.md skills and 5 manifests, does not contain MiniMax Code source code, and can be installed by copying the skill directories, via a local plugin, or through a plugin marketplace, and can also be installed in Claude Code; the original text did not publish evaluation results.

View story:MiniMax Code team open-sources Agent workflow and Skills

Thariq proposed expressing creative ideas to the Agent using sketches

On October 29, 2026, Thariq proposed that he very much hopes to first sketch out animation ideas and character concept drafts himself, and then have the agent refine them, rather than going to the trouble of searching for reference cases. He also mentioned that he has seen artists deeply worried about AI, but he still hopes to use the agent to complete his ideas, and lamented his own lack of this expressive ability. Currently, this statement exists only in one remark by Thariq himself, with no subsequent developments or institutional response.

View story:Thariq proposed hoping to express ideas to the Agent using sketches

OpenRouter announces the return of Royale: Last Agent Standing

OpenRouter announced the return of Royale: Last Agent Standing. According to the information it released, 20 models from 10 labs will compete in 30 matches in a livestream, with only one winner left in the end; the livestream commentary will be provided by ElevenLabs V4, on Thursday, October 29 at 12 PM ET. The same report noted that the cited teaser post stated the scale as 30 models and 20 labs, which is inconsistent with the scale described in the main text. There are currently no subsequent developments.

View story:OpenRouter announced the return of Royale: Last Agent Standing

AgentWeb launches, enabling AI agents to complete website tasks without a browser

On October 29, 2026, AgentWeb released web infrastructure for intelligent agents, enabling AI agents to complete website tasks without a browser by invoking preset actions instead of reading screenshots and clicking buttons. According to a report by Rohan Paul the same day, the company claims it has completed 47/47 real-time tasks across 15 government portals, 7 insurance companies, and 25 sites requiring login such as Amazon and LinkedIn, with a median time of 1.9 seconds and a total cost of 3 US dollars per round. Each mapped task is exposed as a tool, including an input schema, preconditions, permission boundaries, and completion credentials, and can be invoked from any assistant. There are currently no subsequent developments.

View story:AgentWeb goes live: AI agents complete website tasks without a browser

indigo previews end-of-month livestream to discuss evolution toward the Agentic Internet

On October 29, 2026, indigo published a post previewing the theme of this month's end-of-month livestream: "Evolving Toward the Agentic Internet." indigo said they had originally planned to keep it simple, but the content became super rich, and invited readers to check the table of contents and reserve a spot; the livestream will go live on YouTube over the weekend. There is currently no information on further developments.

View story:indigo previews that the end-of-month livestream will discuss evolving toward the Agentic Internet

Tencent, Tencent Cloud and WorkBuddy jointly release AI Agent enterprise solutions handbook

On October 29, 2026, Tencent, Tencent Cloud and WorkBuddy jointly released the 78-page "2026 AI Agent Enterprise Solutions and Implementation Practice Handbook." According to a report by Shao Meng that day, the handbook uses WorkBuddy as the entry point, advancing personal assistants into "digital employees" that enter business systems, with ClawPro, Agent Runtime and TokenHub responsible for provisioning, isolation, models and governance. The handbook proposes a three-stage framework of "able to converse → able to work → able to take up a post," and presents case data including Yuzu's Agent penetration rate exceeding 75%, Youka's MTTR of 5–15 minutes, and operations and maintenance costs of approximately -40%, all of which are self-reported in the text. There are currently no subsequent developments.

View story:Tencent, Tencent Cloud and WorkBuddy jointly release an AI Agent enterprise solutions handbook