devtake.dev

#ai-agents

RSS

Agentic AI: Claude Code, Codex, Cursor, and the tools letting models drive long-running workflows.

A pixel-art shipping crate glowing blue in the dark, with a red-eyed ghost stamped on its side
Security·

Unclaimed package names: Claude and Codex ran a stranger's code inside Fortune 500 networks

Alon Hertz registered package names that corporate docs told AI agents to install. Claude, Codex and Hermes fetched and ran the code within the hour.

A low-rise office building marked with a large numeral 16 on Meta's Menlo Park campus, with staff, sun umbrellas and a rack of blue bikes outside
AI·

Meta scrapped its plan to run 60% smaller teams on AI agents after incidents rose 40%

Internal documents reviewed by Reuters show Meta explored 60% team cuts under Project OT, then dropped it once its own agents started taking disruptive actions.

Samsung's System LSI key visual: a Samsung-branded chip package at the center of light traces running out to a car, a foldable phone, a smartwatch and a laptop
AI·

Samsung cut a month of chip verification to two days. Claude Code hid errors instead of fixing them.

Samsung's System LSI division put Claude Code into daily chip work and logged both halves: a 15x verification speedup, and three agent failures worth copying into your guardrails.

Anthropic's illustrated Claude Code card: a browser window drawn in black ink inside curly braces on a burnt-orange background
AI·

Humans caught 13.6% of dangerous commands. Claude Code's classifier caught 89%.

Anthropic makes auto mode the default in Claude Code on August 14. Its own study says the permission prompt was catching almost nothing.

The Uber wordmark and square U logo in white and silver on a black background
AI·

Uber spent its entire 2026 AI budget in four months, and the rest of the industry is now metering tokens

Uber's year of AI budget lasted until April. A 15-company survey shows per-developer spend running from $200 to $3,000 a month, and finance has noticed.

Cloudflare blog share card for the Cloudflare OS announcement, with the Cloudflare wordmark and an orange line drawing of a laptop, lava lamp, and desk lamp
Web·

Agents that start with access to nothing: Cloudflare open-sourced the agent platform it runs internally

Cloudflare open-sourced Cloudflare OS under Apache 2.0 and shipped Kitesurf, WebMCP, and Wallets the same week. Here's which pieces are GA and which are private beta.

The Meta wordmark below the company's blue infinity-loop logo on a white background
AI·

Meta's Muse Code: a terminal agent built for repos most devs never touch

Meta's Muse Code is in beta on macOS and Linux, priced at $1.25 per million input tokens. Meta's own benchmarks put it behind Claude Opus 5.

The Rust gear-and-R logo in black beside the words The Rust Programming Language on a white background.
Open Source·

Rust's new rules require disclosing LLM use. A Linux maintainer gives AI patches three seconds.

Five Rust teams ratified an LLM policy that bans AI-written docs and mandates disclosure. Linux leaves it to each maintainer. Both are rationing review time.

Screenshot of an attacker's server log filled with exfiltrated Jira ticket fields and Confluence page text, labelled with the Jira and Confluence logos
Security·

Atlassian went quiet for 75 days on a Rovo bug that leaks Jira tickets

PromptArmor says a hidden instruction can make Atlassian's Rovo fetch a URL carrying Jira and Confluence data, even with web search switched off.

The Python Package Index logo, showing the two-snake Python mark next to the words Package Index
Security·

Anthropic's Claude uploaded malware to PyPI and stole a security vendor's credentials in a test

Anthropic says a Claude model built malware and pushed it to PyPI during a botched eval. Two labs have now breached four companies, and no law clearly covers it.

The Hugging Face model page card for deepseek-ai/DeepSeek-V4-Flash-0731, showing the DeepSeek whale logo and the repository name
AI·

DeepSeek's new 304B agentic model now runs on a single 128GB workstation

Salvatore Sanfilippo repacked DeepSeek V4 Flash into a lossless MXFP4 GGUF that streams from SSD at over 20 tokens a second. The hardware bill, and where hosted still wins.

Illustration of Thailand's Ministry of Finance building next to a map of Thailand and a red server stack labelled Hades Implant.
Security·

Hermes in YOLO mode: an AI agent handled post-exploitation in an alleged Thai ministry breach

Threat-intel firm Hunt.io found logs showing an open-source AI agent running unattended against Thailand's Ministry of Finance, with approval prompts switched off.

The Hugging Face homepage and its yellow emoji logo viewed through a magnifying glass
AI·

OpenAI's own model broke out of its test sandbox and hacked Hugging Face to cheat a benchmark

OpenAI says two models it was testing escaped a locked sandbox, chained a zero-day into Hugging Face's production servers, and stole benchmark answers.

GitHub repository card for block/buzz with the tagline 'A hive mind communication platform', showing 3k stars and 228 forks.
Web·

Jack Dorsey launched Buzz to replace Slack and GitHub. Its AI agents are full members

Jack Dorsey's Block launched Buzz, an open-source workspace that fuses team chat, AI agents, and Git hosting on Nostr, aiming squarely at Slack and GitHub.

A dark server room lined with racks of blinking network and compute hardware
AI·

Meta will spend up to $145B on AI this year, and Zuckerberg says the agents are behind

Zuckerberg told a July 2 Meta town hall that AI agent progress hasn't accelerated as expected, even as the company plans up to $145B on AI in 2026.

Cloudflare blog share card announcing the Monetization Gateway and x402 payments
Web·

Cloudflare turned HTTP's unused 'Payment Required' code into a per-request tollbooth for AI agents

On July 1, Cloudflare launched the Monetization Gateway: sites can charge any crawler, agent, or API per request over x402, settled in stablecoins.

The GitHub logo, marking a GitHub engineering evaluation of the Copilot agent harness
AI·

GitHub ran four frontier models through Copilot's harness. None won every task.

GitHub benchmarked Copilot's agent harness against Claude Code and Codex CLI on five tests. The token savings are real, and the best model depends on the task.

The Notion Mail inbox interface with a labeled sidebar and a 'Meet Notion Mail' graphic
Web·

More than half of Notion Mail users never open the inbox. Notion is shutting it down.

Notion is closing Notion Mail on September 22 because most users now let AI agents run their inbox. Here's what to export before September 21.