AI takes more steps on its own, you need to set the boundaries

There is a common thread in this week's How-to stories: AI is becoming less about writing the perfect prompt and more about setting boundaries for something that increasingly acts on its own.

This is clearest in how prompting is changing. Where you previously had to guide ChatGPT step by step, chatbots can now take several steps on their own from one simple message. It is not a big overnight shift, but a gradual development where you as a user have to get used to giving direction rather than detailed instructions.

AI moves into the home

Google goes a step further with the agent CC, which is meant to coordinate the whole family's calendar, shopping and meal plans. You don't need to wait for the future to benefit from it, as there are already concrete things you can ask Gemini to help with at home. The same trend can be seen in the workplace, as OpenAI launches a data plugin for ChatGPT Work so you can query your business data in plain language instead of building database queries yourself.

Control and privacy under pressure

But the more AI is allowed to act on its own, the more important the question of control becomes. A new survey shows that half of Danish companies don't know at all how employees use AI day to day. At the same time, OpenAI's internal Project Lily raises privacy questions when contract workers review real ChatGPT conversations to make the chatbot less sycophantic. It is a reminder that what you write to a chatbot is not necessarily read only by a machine.

Structure beats bans

There is good news, however, for those who fear the solution is to shut down AI entirely. A two-year study from the school world shows that students under a total ban perform worst, worse than both those with free access and those with structured training. The point is the same one that recurs in the story about companies' lack of control over AI use: bans don't work, structure does.

Even when you use AI for something as simple as writing a text, it is about setting boundaries rather than handing over control. A well-known tech figure only uses language models to check his texts, never to write them, and has two simple rules to make sure the text still sounds like him.

Keep an eye on whether your workplace has clear guidelines for AI use, not just bans or a free-for-all. And think twice the next time you share something personal or sensitive with a chatbot. It may be read by more people than you think.