TEMPO.CO, Jakarta - Professor of Artificial Intelligence at the Surabaya Institute of Science and Technology (ISTTS), Esther Irawati Setiawan, highlights a new phase of AI towards greater autonomy. Intelligent systems are now not only responding to users but also capable of gradual thinking. "We have entered the era of Agentic AI, where systems can perform reasoning, planning, and task execution," she said in the Workshop Artificial Intelligence Talent Factory on Friday, April 17, 2026.
According to Esther, this shift is a fundamental change in how humans interact with technology. Initially, AI was dominated by conventional machine learning approaches. This scheme continued through the use of large language models (LLMs) and eventually entered the era of agentic AI.
This development expands the role of AI from just being a tool to becoming a more active work partner. In the workplace, Esther says that the combination of humans and "digital humans" is expected to become an unavoidable new pattern. She mentioned that the use of LLM should be tailored to specific needs because conventional machine learning methods are sometimes more efficient. "Although it is hyped, not all solutions have to use LLM. We need to avoid excessive approaches," Esther said.
Gadjah Mada University (UGM) Vice-Rector, Arief Setiawan Budi Nugroho, emphasized that AI still depends on the quality of the human behind it. "Without a strong basic understanding, AI can produce erroneous interpretations," he said.
Vice Minister of Communication and Digital, Nezar Patria, believes that this change requires an adequate response in terms of talent readiness. According to him, mastery of AI is not only about technical ability but also an understanding of the direction of technological development itself. "The future job market will be a combination of humans and digital systems. This demands us not only to be able to use technology but also to develop it," Nezar said.
The Risks of Imitation Intelligence Evolution
Citing Tempo's premium report: "Evolusi AI Makin Canggih. Apa Risikonya bagi Kita?" (The Smarter AI Evolution. What Are the Risks for Us?), AI trained on text-based data is gradually becoming models that learn from video and real-world simulations. Researchers from Stanford University note that smart systems are continually updated with new obstacles. In other words, AI technology is continuously developed to overcome output failures, biases, vulnerability to cyber-attacks, high energy needs, and high computational costs.
Emu Video from Meta and Sora from OpenAI are two of the most popular examples of multimodal AI approaches. Emu is a text-to-video generator, while Sora creates and edits video based on text commands. In this context, the video output shows how AI assembles information about motion, changes in time, cause-and-effect relationships, and interactions between objects.
Tesla, Inc. is a well-known technology company that uses multimodal AI for its autonomous vehicle systems. The electric car manufacturer continues to promote a video-first training approach based on an AI network that uses real-world vehicle camera data.
Read: Who Is John Ternus? Meet the Next CEO of Apple