AI Is Getting Better at Doing Its Own Chores
Today’s AI news has a quiet theme running through it: automation eating its own homework. From tools that set themselves up to agents that grant themselves permissions, AI systems are increasingly handling the tedious work that used to fall on human developers — and that shift has real consequences for everyone downstream.
Anthropic’s AI Agents Can Now Give Themselves Permission to Use Tools
Managed Agents — AI systems designed to complete multi-step tasks by calling on external tools like search engines or calculators — just got easier to build. Anthropic released version 1.6.0 of its Python SDK this week, and the headline change is automatic tool permissions. Previously, developers had to manually configure which tools an agent was allowed to use, every time they built one. Now the system handles that step itself.
Think of it like the difference between a new employee who needs their manager to approve every software login versus one who can request and receive access automatically based on their role. The agent figures out what it needs and gets it, without anyone holding its hand through the process.
For most people, this change is invisible. But it shortens the path from “someone had an idea for an AI assistant” to “that assistant actually works.” Fewer manual steps in development means products get built faster, and that eventually shows up as more capable apps and services in your hands.
Why this matters: The faster AI agents can be built and deployed, the sooner they show up in the products you actually use.
“Automatic tool permissions in Managed Agents for flexible external tool use.”
Meta Is Using AI to Set Up WhatsApp Business So Developers Don’t Have To
According to TechCrunch, Meta has reportedly launched an AI-powered tool that automates the setup process for WhatsApp Business — the version of WhatsApp that companies use to communicate with customers at scale. Configuring WhatsApp Business has historically involved repetitive, time-consuming tasks: writing message templates, running compliance tests, checking that everything connects properly. Meta’s new tool reportedly handles all of that automatically.
The analogy here is tax software. Nobody enjoys entering the same information into a dozen different forms. Software that pre-fills the tedious parts doesn’t change what you’re trying to accomplish — it just removes the friction so you can focus on the actual goal.
If this works as described, small businesses that want to use WhatsApp to reach customers will have a much lower barrier to entry. Right now, setting up WhatsApp Business properly often requires hiring a developer. A tool that automates the boring setup work could put that capability in the hands of a shop owner who just wants to send order updates to customers.
Why this matters: Lower setup costs mean more businesses can afford to use these tools, which means more of your customer service interactions may soon run through AI-assisted channels.
“AI assistants to automatically handle tedious technical work for WhatsApp Business setup.”
The AI Graveyard Is Getting Crowded — Even for the Big Players
TechCrunch has published a running list of AI projects that quietly failed, and the names on it are surprising. According to the piece, Apple’s Siri improvements arrived far later than promised, and OpenAI’s attempt at an all-in-one app never found its footing. The broader argument is that even companies with enormous resources and talented teams regularly fail to deliver on what they announced.
This matters because the public narrative around AI has often been one of relentless progress. Every week brings new announcements. What gets less attention is everything that gets quietly shelved. The graveyard TechCrunch is building puts that quieter story in one place.
For anyone making decisions based on AI promises — whether you’re a business owner evaluating a new tool or a student thinking about a career — this is a useful corrective. Announcements are not products. Demos are not deployments. The gap between what AI can do in a controlled setting and what it reliably does in the real world remains wide.
Why this matters: Understanding which AI projects fail, and why, is just as important as tracking which ones succeed.
“Even well-funded companies struggle to deliver on their AI ambitions.”
Also Happening in AI
A handful of quieter but useful releases landed this week. OpenAI fixed a bug in its Python SDK — the software library developers use to connect their apps to OpenAI’s models — that was causing problems with how failed requests were retried. Ollama, a popular tool for running AI models on your own computer, released version 0.34.1 and removed the “experimental” label from a file format called safetensors on Apple machines, meaning that feature is now considered stable for everyday use. ComfyUI, an open-source tool for generating AI images, pushed version 0.36.0 with several workflow improvements. On a heavier note, Philadelphia is reportedly weighing whether to allow AI data centers in an industrial neighborhood already dealing with the environmental fallout from previous heavy industry, according to TechCrunch. And MIT Technology Review published a roundtable where researchers from major AI companies debated the question everyone keeps circling back to: whether advanced AI poses a genuine existential risk.
What to Watch
The real story threading through today’s news is who controls AI systems as they become more autonomous. Agents that grant themselves permissions and tools that configure themselves are useful — but they also mean fewer human checkpoints in the process. Watch for regulatory attention to shift from AI outputs toward AI autonomy in the next few months. The Philadelphia data center debate is also worth following: it’s a preview of how communities with existing industrial scars will push back when AI infrastructure comes looking for land.