INDUSTRY.co.id - Artificial intelligence (AI) is no longer at the margins of business in Indonesia. Organisations are exploring how it can improve operations, reshape workflows and create new sources of value, with adoption gathering pace across the economy.
According to UOB's 2026 Business Outlook Study, 74% of Indonesian businesses have adopted AI, although half of these adopters remain at an early stage, with only around a quarter reaching an advanced stage. With 90% planning to increase their AI budgets in 2026, there is clearly no shortage of appetite. The more consequential question is whether organisations can translate that investment into sustained, measurable business value.
That requires a shift in how businesses think about AI. Enterprise value rarely comes from generating a useful response in isolation. It comes from what happens around it, when data is retrieved, decisions are made, systems are updated, exceptions are escalated and outcomes are measured.
This is where AI agents become increasingly significant.
Unlike a chatbot that responds to a prompt, an agent can pursue an objective across multiple steps, interpreting information, selecting tools, taking action and adapting based on the result. That opens opportunities across customer service, IT operations, knowledge management, and other workflows where work depends on a sequence of decisions rather than a single answer.
It also raises a more consequential question. When AI can act, how much autonomy should it have?
Who oversees autonomous agents?
AI maturity should not be judged by how independently an agent can operate, but by how deliberately an organisation decides where that autonomy belongs. Indeed, some processes may be suitable for independent execution but others will require human approval or closer oversight.
This makes governance an operational discipline rather than a final compliance exercise. Organisations need clear boundaries around data access, system permissions, and decision-making authority, alongside sufficient traceability to understand what an agent did and why. They also need mechanisms to intervene when behaviour falls outside expectations.
Trust does not come from assuming an AI system will behave correctly. It comes from designing the conditions under which it can be trusted. As agents become more capable, that means calibrating autonomy to the nature and risk of each task.
The real challenge begins in production
The industry has become remarkably good at demonstrating what AI can do. The harder task is making it work reliably inside a real business.
A pilot can operate within controlled conditions. Production is less forgiving. Data changes, systems behave unexpectedly, business rules evolve, and costs accumulate. An agent that performs impressively in a demonstration can behave very differently when exposed to the complexity of an enterprise.
Scaling AI is therefore not simply a matter of deploying more capable models, but hinges on whether the organisation has the foundations to support them.
A more effective approach would consider production readiness needs from the outset through structured development and testing, controlled deployment, and continuous evaluation. This makes it easier to measure performance against business outcomes, alongside operational and financial signals.
The question is not simply whether an agent can complete a task. It is whether it can do so consistently, within defined boundaries, and at an acceptable level of risk and cost.
Scaling without losing control
As more AI systems move into production, development, governance, deployment, and monitoring can become fragmented across teams and technologies. That may be manageable for a handful of experiments, but it becomes a serious constraint when AI operates across multiple functions and at greater levels of autonomy.
Scaling AI, therefore, requires more than scaling use cases. It requires an operating model that connects development and testing with governance, deployment, monitoring, and intervention.
Organisations also need visibility across AI systems, so they can understand performance, identify problems early, and continuously improve as conditions change. Without that connective layer, businesses risk creating isolated pockets of AI capability that are difficult to govern and even harder to scale.
Indonesia's next AI chapter
Indonesia has already demonstrated strong momentum. Komdigi reported 92% adoption of Generative AI among skilled workers in February 2026, while also making clear that adoption alone does not guarantee productivity or economic value.
The organisations that gain the most from AI will not necessarily be those that deploy the most models or agents, but those that understand where AI can act with confidence, where humans need to remain accountable, and how the two can work together reliably as business conditions change.
That requires deliberate and concerted effort to build the organisational capability to create trustworthy systems that can be measured against meaningful business outcomes and continuously improved over time. The companies that can establish that capability will be better positioned to turn today's rapid adoption into a lasting source of competitiveness as Indonesia enters the next chapter of its AI journey.
