For the past two years, most people have been using AI in the same way: typing questions, waiting for answers, and then copying the results. That pattern is beginning to shift. The latest generation of AI not only answers but also works. It can design steps, execute commands, check its own results, and then deliver something that is complete.
This is called agentic AI. Interestingly, this wave is not just touching programmers. Administrative staff, lecturers, and even small business owners are starting to become "creators" of their own work tools.
The difference is agency, not just assistance
Regular AI assistants stop at text. Agents go further. They are given a goal, break it down into tasks, execute them, and verify themselves whether they are correct. Google introduced the Antigravity platform in November 2025 alongside the Gemini 3 model. This platform, which is a modification of Visual Studio Code, allows agents to see the editor, execute commands in the terminal, and even observe the browser window to test newly created applications. OpenAI, through Codex, is taking a similar direction. The control remains in human hands, but the rough work is taken over by machines.
Building small applications without a coding background
The implications are real for daily affairs. For example, a campus administrative staff member used to have to queue to request a simple tool from the busy IT team. Now, they can explain their needs in plain language, and the agent will draft the application framework, create the initial interface, and add features like a scheduling assistant or an incoming mail recorder. These are not giant applications, but small tools that solve specific problems. This is where the biggest impact lies: thousands of small needs that have long been "neglected" because they were considered too trivial to enter the development queue.
Why this is important for Indonesia
Indonesia aims to prepare around 9 million digital talents by 2030, and the skills gap remains a classic obstacle in many offices and campuses. Agentic AI offers a different shortcut. Instead of waiting for everyone to learn programming languages, the ability to build some things can be democratized through tools that understand human intent. Of course, this is not without caveats. The results of agents still need to be checked, data security issues must not be overlooked, and business logic remains the responsibility of humans. Machines accelerate, not replace, consideration.
The shift from "using AI" to "building with AI" may become the most noticeable productivity differentiator in the coming years. Those who are accustomed to treating AI as a colleague that can execute, rather than just a answering machine, will have an advantage that is hard to catch up with.
References:
- Google Developers Blog – Build with Google Antigravity, our new agentic development platform → developers.googleblog.com
- McKinsey & Company – The economic potential of generative AI: The next productivity frontier → mckinsey.com
- Kementerian Komunikasi dan Digital RI – Program penyiapan talenta digital nasional → komdigi.go.id