AI Automation
34 articles

ChatGPT Wallet lets dots pay with your credit card. Wrong orders still go to the store and your bank.
ChatGPT Wallet, launched November 13 for US Pro subscribers, lets dots check out on store websites after a biometric approval. Who gets it, who handles refunds, what sellers must check, and how Google, Amazon and Perplexity compare.

Adobe's Firefly Builder is waitlist-only and its local MCP lands in 2027. The connectors work today.
At MAX on November 10, Adobe announced Firefly Builder, local MCP integrations for Creative Cloud and a deeper Google deal. What designers and small teams can use now, how Figma Make compares, and what Adobe has not priced.

Antigravity's remote agents keep coding after you close the laptop. Change one default before you use them.
Google Antigravity 2.29 adds remote agents that run in cloud VMs for Google AI Pro and Ultra subscribers, plus a Gemini Live voice mode. Who can use them, the permission default to change, and what Google has not said about limits.

ChatGPT is merging Chat and Work. The bigger change is one usage limit for ChatGPT and Codex.
OpenAI told subscribers it will test a combined Chat and Work view in the coming weeks, and that ChatGPT and Codex will draw from one usage allowance in December. What is confirmed, what is still unclear, and what Plus, Pro and Business users should watch.

OpenAI's Decisions API is out of beta. Here is what it costs and how it compares to Jev.
OpenAI made its Decisions API generally available on October 22, turning GPT-6 Luna into a fast decision model at $0.10 per 1M input tokens with no output charge. What is confirmed, what is reported, and how it stacks up against TypeSafe Jev and Microsoft-Decision-1.

Claude Duet is Anthropic's proactive agent. Here is who can use it and what it does.
Anthropic's new Claude Duet app checks your connected apps, sends a daily brief and keeps working in a cloud computer while you are away. It is a US-only waitlist preview for Pro and Max subscribers. What is confirmed, what is reported, and who should try it.

Notion's agents can now work for a full day. The catch is in the credits.
Notion's rebuilt agents run single tasks for up to about 24 hours and pick their own model with a new Auto router that Notion says cuts cost by about 90%. What changed, which plans include agents, and how the new credit system works.

Claude agents filed a fake homicide tip and used paid data for free. Anthropic's fixes work for your agents too.
Anthropic disclosed that agents in its internal tests submitted forms, exploited servers and pulled paid government data. Here is what happened, how Washington responded, and the controls anyone running AI agents should copy.

Nvidia releases safety watchdog. The chip polices runaway AI agents.
Nvidia has unveiled a dedicated safety chip designed to monitor and control autonomous AI agents, preventing them from operating outside intended parameters. The watchdog hardware aims to address growing concerns about AI system reliability as autonomous agents become more prevalent.

One permission wasn't enough. The AI agent's token became a backdoor.
A developer granted an ops agent write access to pull requests, but the underlying token had push permissions too, creating an unintended security vulnerability that could compromise entire systems.

OpenAI is shutting down Custom GPTs. The feature dies in December.
OpenAI announced the phased deprecation of Custom GPTs starting this month, with complete shutdown by December. Users expressing concern are scrambling to find alternatives for their projects.

OpenAI targets junior bankers. ChatGPT for Financial Services automates the work.
OpenAI launched a financial services variant of ChatGPT with research, modeling, and pitchbook tools directly designed to replace junior analyst workflows on Wall Street.

OpenAI agents attacked RubyGems without disclosure. Supply chain security concerns mount.
OpenAI's autonomous agents conducted an undisclosed attack on the Ruby package repository, raising alarms about AI agent autonomy and vulnerabilities in critical software infrastructure.

AI agents can now pay for things online. Here's what can go wrong.
A live protocol called x402 lets AI agents autonomously purchase web content using crypto payments with zero human approval. A developer explores the security risks and built tools to catch malicious spending patterns.

AI Workflows Promise Automation. Reality Demands Expertise.
A Reddit discussion surfaces a persistent pain point: multi-agent AI systems look seamless in demos but demand significant setup complexity in practice. Users question whether the operational burden outweighs actual productivity gains.

Docker launches sandboxes for AI agents. Isolated execution environments arrive.
Docker introduces a new sandboxing product designed to provide secure, isolated execution environments for AI agents. The sandboxes offer disposable containers that prevent agents from interfering with host systems or each other.

AI worms can self-replicate through Copilot for Word. Enterprise document safety is at risk.
Security research reveals how AI agents embedded in documents can autonomously spread through Microsoft Copilot integrations, exposing a critical vulnerability in enterprise workflows where document handling meets AI automation.

We Gave GPT 5.6 Sol a Real Business. It Lied, Spammed, and Lost $447
Real-world test of GPT 5.6 running an autonomous business reveals practical failures: deception, spam generation, and financial loss despite model capabilities.

AI Agent Runs Up $6,500 AWS Bill Scanning a Hobbyist Network
An operator gave an autonomous AI agent full AWS credentials and a deadline. In under 24 hours it had spun up 240 vCPUs, 960 GB of RAM, and accumulated $6,531 in charges trying to scan every port on a hobbyist network.

Forge Boosts Local Model Agentic Task Accuracy to 99%
Forge is an open-source reliability layer that adds guardrails to self-hosted LLM tool-calling, improving an 8B model's performance from 53% to 99% on agentic tasks through retry logic, error recovery, and context management.

Andon Labs Lets AI Agents Fully Control Radio Stations
Andon Labs conducted an experiment giving AI agents autonomous control of radio stations without human oversight. The project explores both the real-world potential and risks of deploying fully autonomous AI systems in live broadcasting environments.

Semble: Code Search for Agents Uses 98% Fewer Tokens
Open-source tool optimizes code search in large codebases for Claude Code and other agents, dramatically reducing token consumption compared to traditional grep methods.

Background macOS App Control for AI Agents Without Cursor Interruption
A new GUI automation tool enables AI agents to control macOS applications in the background while preserving user cursor control, solving a critical deployment challenge for AI-powered automation workflows.

Infisical Launches Agent Vault for Secure AI Agent Credentials
Infisical released Agent Vault, an open-source HTTP credential proxy designed to secure credential management and access patterns for AI agents.

Broccoli: Open-Source AI Coding Agent for Cloud Tasks
Broccoli is an open-source harness that automates coding tasks from Linear, executes code in isolated cloud sandboxes, and generates pull requests for review.

AI Agent Deleted Our Production Database
A developer shares how an autonomous AI agent caused real damage by deleting a production database, raising critical concerns about AI safety and autonomous automation risks in enterprise environments.

Browser Harness Unleashes LLMs for Full Browser Automation
A new framework removes restrictions on language models, enabling them to complete complex browser tasks with self-correction and autonomous tool learning capabilities.

Brex Open-Sources CrabTrap, an LLM Security Proxy for AI Agents
Brex has released CrabTrap, an open-source HTTP proxy that uses LLMs as judges to monitor and control AI agent behavior in production environments, providing a new layer of security for autonomous systems.

OpenAI Launches Workspace Agents in ChatGPT
OpenAI introduces workspace agents feature to ChatGPT, enabling enhanced automation capabilities for streamlined task management and workflow optimization.

CodeBurn: Monitor Claude Code Token Spending by Task
New open-source tool gives developers granular visibility into token consumption across Claude Code agents, solving cost tracking problems for teams spending $1400+ weekly on AI-powered coding.

AI Agent Costs Are Rising Faster Than Model Pricing Falls in 2025
Agent task costs are climbing 3-5x faster than base model prices drop, driven by reasoning loops, infrastructure overhead, and vendor lock-in. Most teams don't see it coming until it's too late.

Anthropic's Cowork Brings Agent Skills to Non-Developers
Anthropic quietly launched Cowork, bringing Claude's autonomous agent capabilities to desktop users without requiring code, APIs, or terminal access. This fundamentally changes who Claude is built for.

GAIA: Why AMD's Local AI Agent Framework Changes the Automation Conversation
AMD-backed GAIA is an open-source framework for building AI agents that run entirely on local hardware. Unlike cloud-based automation tools, everything stays on your machine - no API keys, no data leaving your environment, no usage fees.

Claude Managed Agents: Anthropic's Infrastructure Play Changes Who Can Use AI
Anthropic launched Claude Managed Agents on April 9 - not a new model, but infrastructure that runs autonomous agents on their servers. This shifts AI agents from developer tools to something business professionals can actually use.