AI Agents news and analysis on Web Pulse: 49344 curated articles tracking the latest AI Agents developments, tools, and research, updated continuously from vetted sources.
Agent37 published a pricing comparison on October 10, 2026, claiming a 2 vCPU, 4 GB AI agent sandbox costs $4.76 for 730 running hours, versus $248.78 at Vercel, $145.28 at Cloudflare Sandbox and $120.89 at Daytona and E…
A malware analysis lab can be built from a Windows VM, a REMnux VM or container, an AI agent, and a remote AI model, with the Windows VM and REMnux connected over a host-only virtual network that has no internet access, …
DeepSeek published an article outlining a "Super Agent" design in which a manager agent coordinates other agents, and the author reports testing a similar architecture built from coding agents such as Claude Code and Cod…
Collinear AI released CWE-bench v2, a public set of 100 security tasks covering 91 different CWEs, in which an agent must find and fix a vulnerability in a real open-source repository without breaking existing functional…
MotherDuck launched Agent Chat, a natural-language interface built into the MotherDuck UI that lets a native agent write and execute SQL and Python, build data pipelines through MotherDuck's Flights Python runtime, and c…
Entire launched Trails, a review loop that places eval scores at the top of each page and recalculates them on every push, with each review agent running independently in a cloud sandbox and returning a score and rationa…
A developer tested three long-standing rules in framesmith's CLAUDE.md by deleting each one in a throwaway git worktree and running a headless task with `claude -p "<task>" --permission-mode acceptEdits`, finding one rul…
Telnyx Inference has made DeepSeek V4.1 Flash 1017, the post-training checkpoint of DeepSeek's open-weight DeepSeek-V4.1-Flash model, available on Telnyx-owned GPUs under the model ID deepseek-ai/DeepSeek-V4.1-Flash-1017…
Collinear AI published CWE-bench v2 on November 3, 2026, a public benchmark of 100 security audit-and-patch tasks spanning 91 CWE types and eight programming languages, with Gemini 4 Argon leading 11 evaluated models at …
GitBook reported that an attacker gained "unauthorized access to some of our infrastructure credentials" at 19:00 CET on November 2, 2026, in what the documentation platform described as a "sophisticated, highly targeted…
Binance Co-CEO Richard Teng said on CNBC's "Squawk on the Street" on November 3 that the roughly $6.3 billion theft from Binance wallets on October 29 began in June with a fake developer tool and in September with fake r…
Vercel added a `vercel doctor` command to its CLI that checks the CLI version and install, login and current team, project link, `vercel.json` configuration, and active incidents from the Vercel status page, reporting ea…
A developer built a portfolio website for a content creator by using Spidra to pull structured data from the creator's public Instagram posts and reels and feeding it to an AI coding agent, according to a first-person ac…
Anthropic's Claude Code sandbox, enabled via the /sandbox command, confines bash command writes to the working directory and filters outbound traffic by hostname, but leaves credential files such as ~/.ssh and ~/.aws/cre…
Agents using GitHits read every dependency at the lockfile-pinned version in all 192 runs, compared with 94.8% of runs for agents that cloned dependencies themselves and 66.1% for agents starting from a prebuilt snapshot…
Tencent's Hy4 model is now available on Vercel's AI Gateway at 50% off through December 31, 2026, under the model name `tencent/hy4`. Hy4 is the first Hy model with vision, supports reasoning, tool calling, structured ou…
Momentic launched Mo, a QA agent that autonomously plans, explores and bug-bashes controls and user flows in web, iOS and Android apps, and in the vendor's own 60-minute test on four open source apps with 1,396 human-lab…
Prod-code 0.3.49, released at tag v0.3.49 (commit 61aae41) and archived on Zenodo as 10.5281/zenodo.23905176, adds an experimental MCP tool called code_workflow that composes multiple code reads into a single request, ru…
A developer published a Bash function called `confine` that uses the Linux sandboxing tool `bwrap` (bubblewrap) to run AI code agents such as Claude and Codex under kernel-level filesystem isolation, after concluding tha…
A new preprint evaluating AI agents on the IUCN Red List assessor exam found that open-weight models running on local hardware now clear the 75% pass mark, with Qwen3.8 Flash self-hosted on a desktop Mac Studio reaching …