A new era in artificial intelligence

From assistants that answer our questions to agents that get work done

Asking AI to write an email, summarize a long text, or answer questions about code are now familiar uses. Now technology companies are focusing on a broader goal: AI agents that plan a task, use the necessary tools, and proceed by checking the result.

Google and Anthropic’s 2026 announcements offer concrete examples of this direction. So what do these developments mean for everyday use?

What should users in Türkiye pay attention to?

This article was prepared based on official announcements reviewed as of September 16, 2026.

Access to the features in the announcements may vary by country, account, and subscription.

What is an AI agent?

When you ask an AI assistant, “How do I edit this file?”, you can get an explanation. An agent, however, when given the appropriate tools and permissions, can examine the file, make edits, and check the resulting output.

The key difference here is that it can follow multiple steps of a task. But its ability to do this depends on the tools offered by the application you use and the permissions you grant. Not every chat window is an agent with access to your computer.

For example, a task can be described like this:

“Review the product images in this folder, list the ones whose dimensions are unsuitable, and show me your suggestions before making changes.”

In this example, the goal, scope, and stopping point are clear. That is exactly a good starting point for using agents.

Google adds task tracking to search

In its May 19, 2026 I/O announcement, Google introduced “information agents” for the search experience. In the approach the company described, the user specifies a topic and conditions they care about; the agent then tracks developments and sends relevant updates.

One of the examples in the announcement is tracking home listings that match certain conditions. Google stated that this feature would first be offered to Google AI Pro and Ultra subscribers. The same announcement also included plans to create personalized interactive tools and small apps. Google’s official announcement

Such an approach could change the research we do over and over. Instead of typing the same conditions and searching every day, it may become possible to define the conditions once and be notified when there is a meaningful change.

However, an announced feature and a feature available in your account are not the same thing. In particular, access in Türkiye should be checked not from the headline of the announcement, but from the current availability information for the relevant product.

Multi-step work stands out on the Claude side

For Claude Sonnet 5, introduced on June 30, 2026, Anthropic highlighted capabilities such as planning, using tools like a browser and terminal, coding, and carrying out multi-step tasks. These are the company’s own product descriptions and evaluations; they do not guarantee the same success on every task. Anthropic’s official announcement

This direction is especially noteworthy for software developers. Beyond explaining a bug, ways of working that examine the relevant files, suggest changes, and proceed based on test results are gaining importance.

Similar logic can be applied to office work: gathering information from multiple documents, flagging inconsistencies, and preparing a draft ready for review. What is valuable here is not just producing fast output; it is being able to show what information the output is based on.

Where can you start for everyday use in Türkiye?

Without waiting to access all the new features, you can start with small tasks whose results are easy to check, to the extent that the tool you use supports them.

For students: Preparing a study plan from your own lecture notes, having difficult concepts explained with examples, and generating review questions.

For small businesses: Organizing product lists stripped of personal information, finding inconsistencies in descriptions, and classifying incoming requests.

For content creators: Comparing sources on a topic, producing a draft article, and listing claims that need verification.

For those interested in technical work: Examining error logs, having code changes explained, and preparing test scenarios.These are usage suggestions. To measure success, asking a simple question in every task is enough: “How will I check the resulting outcome?”

Describe the task clearly for better results

Instead of a broad request like “Research it for me,” specify the result you expect and the boundaries:

“Compare three tools available in Turkey from their official sources. Write their Turkish support, free plan limits, and the date you checked. Do not guess information you cannot verify. Present the result in a short table.”

If an action is to be taken, also clearly define the authority:

“First prepare the suggestions. Do not delete files, make purchases, or send messages on my behalf.”

Especially in tasks involving more than one step, these boundaries make it easier to evaluate the result.

Is AI necessary for every task?

Choosing the tool according to the task itself may be more practical. While structuring a piece of writing requires interpretation and context, resizing an image to specific dimensions is a direct conversion process.

For example, to change the dimensions of your social media image, you can use Image Studio, and to see the changes between two texts, you can use the Compare Text tool. You can involve AI in the process to determine what you want to change or to evaluate the resulting output.

When evaluating new tools, the most useful criterion—before the model’s name or a flashy promotional video—is the result it delivers in your own work: less repetition, easier checking, and time genuinely saved.