November 2025 marks a significant shift in how developers and creative professionals interact with AI. Google released Antigravity, an agentic IDE built on Gemini 3 Pro, while OpenAI expanded Codex with support for autonomous agents that can execute programming tasks end-to-end. For Indonesia, this trend is no longer a curiosity but a wave that has begun to sweep through academic and office spaces.
The Difference Between Agentic AI and Regular ChatGPT
Most AI users in Indonesia are still familiar with the chatbot model: send a prompt, receive an answer, copy to a document. This pattern has been very helpful, but it still requires humans as manual operators. Agentic AI changes this pattern. Agentic systems are given a goal, then plan their own steps, execute through access to various tools, and validate the results before delivering them to the user.
In Google Antigravity, this pattern is evident. The editor view functions like a regular VS Code, but the manager view allows one developer to dispatch multiple agents simultaneously, working in parallel across different workspaces. Each agent produces an Artifact: task list, screenshot, or browser recording, so the results can be verified without having to read raw tool calls.
Codex and Its Complementarity
OpenAI Codex, in the updated 2025 release, is not just an autocomplete for code. Codex now performs tasks such as debugging based on complete logs, multi-file refactoring, and generating test suites. When combined with agentic IDEs like Antigravity that already have native browser integration, Codex becomes the component that handles the pure programming layer, while Antigravity orchestrates the entire workflow.
The Context in Indonesia: Academic and Office
In academic environments, agentic AI is highly relevant for research management. From compiling literature reviews, scraping data from open journal portals, to preparing drafts for statistical analysis, everything can be accomplished with a combination of prompts and validation. Lecturers and researchers who master agentic workflows will save dozens of hours that would typically be spent on repetitive technical tasks.
In office settings, the pattern is similar but with a different focus. Operational teams that need to process hundreds of forms, compile monthly reports, or prepare regular presentations can delegate many technical steps to agents. What remains for humans is decision-making and quality control, two aspects that actually add significant value to professional labor.
Adoption Challenges
The adoption of agentic AI in Indonesia faces three main challenges. First, a steeper learning curve compared to chatbots, as users need to understand concepts like tool calling, plan-execute-validate, and context window models. Second, the computational costs are still predominantly in dollars, although many local alternatives are starting to emerge. Third, governance: when agents can execute shell commands and access files, access control and audit logs become crucial.
For professionals looking to delve deeper, structured training is the fastest way to understand this new stack without having to trial-and-error alone. Starting with the two main tools, Google Antigravity and OpenAI Codex, will provide a strong foundation for further exploration into the rapidly evolving agentic AI ecosystem.
References:
- Google Developers Blog – Build with Google Antigravity, our new agentic development platform → developers.googleblog.com
- OpenAI – Codex: AI Coding Assistant for Engineers → openai.com
- Kementerian Komunikasi dan Informatika RI – Peta Jalan Kecerdasan Artifisial Indonesia 2020-2045 → kominfo.go.id
- Wikipedia – Google Antigravity → en.wikipedia.org