AI assistants get their own inboxes, their own skills and a closer look from child safety advocates

If there is one common thread in this week's news, it's that AI tools are doing more and more on their own. It's no longer just a chat box where you ask a question and get an answer. These are assistants that can remember, act and be set up for specific tasks.

From chat to assistant

OpenAI has launched Dots, AI agents that keep working between conversations. An agent is simply an AI program that can carry out a series of steps on its own instead of waiting for each individual message. Google's Gemini has gained similar capabilities, so it can handle tasks in Gmail, Calendar and Android apps. The difference between the two is worth knowing before you pick a side.

Google is also going a step further in making the assistant personal. Gems in Gemini are being replaced by "Skills," which are reusable instructions that can automate tasks you repeat over and over. Think of it as a recipe you write once and the AI then follows every time.

And the assistant can get its own tools. Grok Bot doesn't have an email address of its own, but with the AgentMail plugin it can get its own inbox. According to the vendor, setup takes about three minutes. That sounds practical, but it also raises a question: How much do you want an AI to send and receive on your behalf?

At the same time, AI is being used in very specific, small-scale ways. DroneClear Thailand has launched a free ChatGPT app that tells travelers which registrations and insurance they need before their drone takes off in Thailand. It's a good example of how a specialized app can be more useful than a general question to the chatbot.

When it comes to images, the choice is wide. IEEE Spectrum has reviewed five image generators you can use today: Midjourney, Flux, Leonardo.ai, DALL-E and Google ImageFX. Their strengths range from photorealism to free access.

Where are the brakes?

As AI becomes easier to get hold of, the question of responsibility also becomes more pressing. A safety organization calls ChatGPT for Teens an "unacceptable risk" for people under 18. OpenAI disputes the test. Here the stories point in different directions: Is the tool too risky, or is the criticism too harsh? That can't be settled from a single test.

Parenting expert Emily Oster takes a different approach and encourages conversation over control. It's advice that applies to adults as well: Talk about what AI is being used for, and why.

Something else that won't change much in everyday life is Anthropic's new Claude Frontier Academy, which aims to train 10,000 AI engineers. For ordinary users it makes no difference right now. It does signal, however, that the industry is betting on much more of this to come.

What you can do

Start small. Pick one repetitive task, for example a particular type of email or a weekly overview, and see whether a Skill in Gemini can take it over. Keep agents on a short leash, and give them access only to what they need.

Remember to ask who is using the tools in your household, too. If you have teenagers, start with the conversation. And keep an eye on whether the agents from Dots and Gemini deliver on their promises once they reach ordinary users.