Closing the AI Adoption Gap in Indonesia
Like many people, my first experience with AI was through ChatGPT, used mostly as a Google Search replacement. Over time, it became more central to my work — helping with brainstorming, refining essays, structuring routines and goals, even asking about health symptoms. But I knew this experience was not representative of the Indonesian market.
As I spoke with people outside the tech and startup ecosystem, I encountered different reference points for “AI” in Indonesia. One example was Cici (now Dola), an app with a chat interface similar to other popular LLM chat platforms, but packaged with a friendly animated character and localized interactions. This was the app that house helpers, construction workers, fitness coaches had heard of and used. Not ChatGPT or Claude.
With Dola, people ask about the weather, what to cook for dinner, or how to structure a workout. The use cases are simple, the interface is not intimidating, and the trust barrier is low.
In Silicon Valley terms, this would be dismissed as an “AI wrapper” (I’m guilty of this too), but does this framing miss the point?
In Indonesia, what looks like a wrapper can be the product
Without quoting any usage data, the adoption on the ground tells a story that is different from what is covered in the media: consumer AI in most of Indonesia is different from the Silicon Valley narrative of productivity tools, vibe coding, or cutting-edge healthcare applications.
What is often dismissed as a “wrapper” is, in practice, the product. The interface, tone, language, and distribution channel are not simply superficial layers on top of intelligence, because they are what make the intelligence usable in the first place.
AI awareness is still limited to a small share of the population. It is disproportionately concentrated among mostly white-collar workers, younger users, and in tier-one cities. This implies two things: baseline literacy will continue to rise, and that the long-term ceiling for adoption is much higher than current usage suggests.
At the same time, this creates a wide gap between those who already use LLM tools for productivity gains and those who don’t. In the near term, a lot of the value in Indonesian consumer AI will come not from pushing the frontier of model capability, but from builders acting as a bridge — closing this gap by making AI more accessible, familiar, and trustworthy, before adoption naturally catches up.
The AI Adoption Gap
I think about this through a framework I call The AI Adoption Gap — the gap between what AI systems are capable of and what people are able and willing to use in their daily lives.
The gap closes from both ends:
- User fluency grows through exposure and repeated use.
- Technical depth becomes more accessible through better design, familiar interfaces, and trusted distribution.
Both move at different speeds, and shaped by different forces.
Dola is a useful case study about correctly reading which phase the Indonesian market is in and building for that phase.
When the gap is wide, the winning move is not to optimize for capability, but to make AI accessible, familiar, and trustworthy enough that people use it at all.