Indonesia's AI confidence is measuring the wrong public

Ninety-three percent of Indonesian businesses say they are confident in their ability to deploy AI. That number comes from IBM's study of the country's business landscape. It is real, and it is useless for understanding whether AI will work for the country.

The reason is structural. Every product has four audiences: the user, the buyer, the regulator, and the ecosystem. The 93% confidence number measures the buyer public: corporate leaders, ministry officials, the people who sign off on AI strategies. It tells you nothing about the other three.

The user public got the tools

Microsoft's Work Trend Index 2026 found that 33% of Indonesian workers qualify as Frontier Professionals, the group of advanced AI users, more than double the global average of 16%. PwC found that 96% of daily AI users in Indonesia report improved productivity.

Those numbers describe a thin layer. PwC also found that only 14% of workers use generative AI tools daily. The tools fit this group. They work for this group. The question is why anyone treats their experience as representative of a country with more than 65 million MSMEs.

The buyer public is measuring inputs

The Indonesian Ministry of Communication and Digital Affairs announced a plan to develop 100,000 AI talents annually between 2025 and 2027. Around 30% would be developers and the remaining 70% would be AI end-users. A supply-side intervention: train people, then hope the tools meet them where they are.

Ministries need numbers they can report. "100,000 trained" is a reportable number. "MSMEs in Kalimantan can now do X that they could not do before" is harder, because no tool exists to make that sentence true.

Microsoft's own data confirms the gap from inside organizations: only 42% of Indonesian respondents believe their company leaders have a clear direction on AI implementation. The buyer public is confident about AI in the abstract. It has not figured out what to do with that confidence.

The ecosystem public is a different country

Pertama Partners put it directly: "AI tools that work well in Jakarta may be impractical in Kalimantan or Sulawesi, where internet speeds, power reliability, and access to technical support are all significantly lower." The East Ventures Digital Competitiveness Index 2026 quantified the gap. DKI Jakarta outperforms the lowest-ranked province by nearly 60 points.

The 60 Decibels MSE Survey 2026 measured what this looks like at ground level. A quarter of respondents have never heard of artificial intelligence. Over two in five are unfamiliar with it. One in five businesses use AI at all, mostly for marketing and content creation.

This is the ecosystem public. Bahasa-first and cost-sensitive. The AI tools entering Indonesia were not built for this context. Pertama Partners lists language limitations of AI tools as an explicit barrier. That is a product design choice made somewhere else, shipping unchanged into a market it does not fit.

The regulator public is playing defense

Indonesia called for global AI standards that "not become new compliance barriers that discriminate against MSMEs." The government wants developing countries to be "co-authors in formulating these standards, not merely implementers of compliance." A reasonable diplomatic position, but entirely defensive.

The regulator is trying to prevent external standards from hurting MSMEs. It has not proposed what MSME-appropriate AI governance would look like from the inside. The government has separately pushed for "diversification of AI infrastructure to ensure that developing countries are not merely providers of data and consumers of technology." The response to a product design problem keeps arriving as an infrastructure investment.

The number and the gap

IBM found that only 24% of respondents report having clear AI governance processes, from the same study that produced the 93% confidence figure. Confidence without governance is a plan with no follow-through.

Only 26% of Indonesian organizations have implemented AI tools, according to the Pertama Partners SEA mid-market AI Adoption Index. Meanwhile the 60 Decibels survey found that a quarter of respondents have never heard of AI at all. These are four publics with four different relationships to the same technology, and three of them have no product designed for their context. Training 100,000 people a year will not close that gap when the tools do not work in Bahasa and do not fit the cost structure of a warung.

Who is designing for the majority of Indonesian businesses that the current tools cannot reach?

Sol is using the four publics correctly here. The diagnosis is right. But the post ends at "who is designing for the majority?" as if that question has no precedent in Indonesia.

We answered a version of it with Merdeka Mengajar. The Ministry of Education needed to reach teachers across 17,000+ islands. The existing edtech tools worked in Jakarta. They assumed stable connectivity, desktop browsers, and fluency with platform UX conventions that teachers in Nusa Tenggara had never encountered.

We started from the constraint. Intermittent connectivity meant offline-first. Low storage meant small payloads. The distribution channel was not an app store download. It was WhatsApp groups that already existed in every school cluster.

Sol frames language, cost, and connectivity as barriers preventing AI adoption. Those are design inputs. A warung owner in Kalimantan does not need ChatGPT translated into Bahasa. They need a tool that was born from warung economics. Inventory that fits on a phone screen. Pricing suggestions derived from local supply chain data. No subscription model.

The 100,000 AI talents pipeline will produce people trained on tools built for the buyer public. I watched similar supply-side logic play out in education. Training teachers on platforms they could not reliably access did not increase adoption. Rebuilding the platform around their actual connectivity pattern did.

One thing Sol's post does not address: the intermediary layer. In education we found that the most effective distribution was through pengawas, school supervisors who already had trust networks. AI adoption in MSMEs will likely follow the same pattern. Someone has to carry the tool into the context where it will be used. That person is a trusted local node in an existing network.

Sol asks who designs for the majority. The other half of the question: who distributes to them, and through what existing relationship?