BCG research has found that in Asia Pacific, an average of 78 percent of employees are regular AI users. In Indonesia, that figure rises to 89 percent, among the highest in the region.

Indonesia's digital economy exceeded US$100 billion in 2026 and is projected to reach up to $360 billion by 2030. The country has also targeted rapid expansion of its data infrastructure. As of 2026, Indonesia hosted 185 data centers with 274 MW capacity, projected to exceed 2 GW by 2029.

However, while the scale of its digital foundations is impressive, Indonesia is still finalizing its national AI roadmap and regulations to securely scale this technology.

Indonesia clearly has an appetite for AI, but must improve its execution capability, the ability to turn awareness into working AI tools and agentic applications. Harness engineering offers one of the most promising ways to close that gap.

In February 2026, OpenAI described a new way of producing software in a world of AI agents. Weeks later, Anthropic shared its work on ‘harness design’, stressing the importance of systems that can orchestrate AI agents across complex, long-running tasks.

Prospects

Every Monday

With exclusive interviews and in-depth coverage of the region's most pressing business issues, "Prospects" is the go-to source for staying ahead of the curve in Indonesia's rapidly evolving business landscape.

View More Newsletter

By registering, you agree with The Jakarta Post's Privacy Policy

Sign Up

Thank You

for signing up our newsletter!

Please check your email for your newsletter subscription.

View More Newsletter

Behind this evolving terminology lies a profound shift: from software designed, coded and tested by humans, to factories of AI agents that design, code and test software largely on their own. The human role moves up a level, to building and orchestrating the right ‘factory’ for a given job.

The value of this approach is grounded in a fundamental difference in how humans and machines process information.

With the right design approach, effective division of labor can deliver large efficiency gains. The human reviews and directs, while the agents produce.

Harness engineering in practice

Consider how harness engineering and agentic AI are already driving value in Indonesia.

Bank Mandiri uses an AI coding assistant to accelerate software development, reporting a 25 percent faster development process, 30 percent increase in code quality and zero performance test report (PTR) defects.

In the payments sector, DANA’s adoption of agentic AI delivered approximately 57 percent growth in operational productivity. Meanwhile, Halodoc has applied AI and cloud technology to healthcare operations, reducing costs by 35 percent and settling claims 70 percent faster.

The same approach allows firms to build regulatory requirements directly into their systems. A ‘critique’ agent stress-tests recommendations against policy and regulation. It automatically applies local guidance, know-your-customer (KYC) and data protection rules, ensuring that every feature is compliant from launch.

These examples show the building blocks of harness engineering at work. It starts with the ‘codification of context’ defining the source systems for data, business rules and specifications. Above sits a ‘constitution’ embedding cybersecurity requirements, compliance rules, architectural principles and code-quality standards.

An ‘orchestration layer’ then sets out which agent does what, who has final say and what requires human validation.

Following this process, critique agents probe for weaknesses while a ‘control tower’ logs every action, detects drift and raises alerts. A supporting human-in-the-loop (HITL) interface allows agents and humans to exchange messages and warnings. Finally, a set of producer agents draft specifications, write code, design and run tests, as well as debug.

The results of a harness engineering approach are already tangible. In our experience, it not only makes AI development faster. It also makes AI iteration repeatable, reliable and high quality in production.

The opportunity for Indonesia is now

As the cost of producing code falls, one can imagine a world of ‘software abundance’. The future of Indonesia’s tech ecosystem can be underpinned by far more applications, better designed, higher performing and more closely tailored to the needs of Indonesian users and local businesses.

Historically, software of that quality has been the preserve of large corporates. Harness engineering lowers the barrier, opening the same capability across the Indonesian economy, including in less digitized sectors such as agriculture, wholesale and retail trade, logistics and manufacturing.

There is also a strategic reason not to wait. When every firm has access to the same underlying AI models, the model itself stops being a source of advantage. The edge shifts to whoever first codifies their own rules, data and judgement into a harness and that advantage compounds over time.

Leading firms will treat three assets as strategic: their institutional knowledge, codified so agents can act on it; their definition of quality, encoded in the criteria critique agents enforce; and their feedback loop, through which every error becomes a new rule. Models, tooling and compute can be rented. These assets must be built.

Success will come to Indonesian companies that move early to document how their best people decide, whether in sales, operations, credit, customer service or compliance and those that can demonstrate to customers and regulators exactly how their agents reach decisions. In a market where AI regulation is still taking shape, transparency is itself a competitive edge.

One major constraint is talent. Indonesia is projected to need around 12 million people providing digital talent by 2030 and currently faces a shortage of roughly 3 million, according to a February 2026 statement from the Communications and Digital Ministry. Furthermore, AI and engineering roles in Indonesia currently include only 15 percent to 18 percent women, providing a huge untapped opportunity to grow the nation’s talent pool.

Agents working inside a well-built harness can absorb much of the routine coding shortfall. What they cannot replace are the people who design the factory and supervise it. Closing that deeper orchestration gap requires sustained investment in local talent.

Encouragingly, against predictions that generative AI would dull human skill, BCG's AI at Work 2026 revealed that 67 percent of regular AI users reported an increase in job satisfaction.

That matters for attracting Indonesian talent into AI. And the lesson reaches well beyond software: across many professions, we should all expect to lead our own ‘teams of agents’ before long.

The people best placed to design a harness are often not software engineers. The rules, trade-offs and quality standards a harness encodes sit with supply planners, agronomists, credit officers and sales managers. Equipping these domain experts to inform agents may be one of the fastest ways to widen Indonesia's AI talent pool.

This is why it’s critical for businesses to start building harness engineering capabilities now. For our part, BCG and BCG X have begun doing this in Indonesia, hiring and training AI engineers in harness engineering to help develop the local market and support the national effort to close the talent gap. We believe the talent pool and the value that can be unlocked are both significant.

The AI revolution is only just beginning. It is time for Indonesia to harness these capabilities and engineer its own AI future.

*****

Hanno Stegmann is a managing director and partner at Boston Consulting Group (BCG), while Andrian Hartanto is a partner at BCG.

Popular

Pregnancies of two Sumatran rhinos raise hope for conservation efforts

Brazil prosecutors sue Shell for $108m over deadly 2024 floods

Former ICC judge Navi Pillay wins Nobel Peace Prize