Beyond the chatbot what agentic AI can actually do for Malaysia of public service

By Dr. Behrang (Hani) Parhizkar, Group CEO of Chamrun Digital

Most government AI projects I am asked to review start in the same place: a chatbot on the front page of a portal. It answers questions and takes some pressure off the call centre. It rarely changes how the work behind the counter gets done. That is the gap agentic AI is built to close.

A chatbot responds. An agent acts. Given a goal, an agentic system can break the work into steps, pull information from several systems, check it against rules, draft an output and pass it to a person to approve. Instead of telling a citizen which documents a licence needs, an agent can check the documents that were submitted, flag what is missing and prepare the case for an officer’s decision.

For Malaysia’s public service, that difference matters more than any model benchmark.

The real bottleneck sits between agencies

Since 2023 I have spent much of my time training senior public officers and building working prototypes with them. The most consistent lesson is that the hardest problems sit between systems. Data lives in one department, the workflow runs through another and the decision needs a third. Officers become the integration layer, copying information from screen to screen.

Agentic AI can act as a connector across those silos. It does not need every system to be replaced. It needs secure, permissioned access to the systems that already exist, clear rules about what it may do and an officer who stays accountable for the decision.

Where agents beat chatbots

Three kinds of public-sector work suit agents well.

1. Checking and preparing. Form checks, licence assessments and eligibility screening follow rules that can be written down. An agent can do the first pass in minutes and leave the judgement to the officer.

2. Sensing and planning. Food security is a good example. Bringing crop, supply and market data into one view lets planners see shortfalls earlier and plan imports with better information.

3. Answering at scale. Multilingual voice and chat agents can handle routine citizen enquiries around the clock, verify reports and route urgent cases to the right team.

In each case the agent is not the decision-maker. It shortens the path to a good decision.

Sovereign and secure by design

Governments are right to be cautious. Citizen data, policy drafts and security information cannot go into whichever tool happens to be convenient. That is why I argue for sovereign AI: systems that run on approved infrastructure, use models suited to the task and keep data inside national and agency boundaries.

In practice, that often means smaller, domain-specific models instead of one giant general model. It means retrieval from the agency’s own documents and a full audit trail of every action an agent takes. Security is not a feature added at the end. It shapes the design from the first workshop.

Build with the people who will use it

The most successful prototypes I have seen were not designed for officers. They were designed with them. When senior officers sit in the room while the agent is built, three things happen. They find the edge cases no consultant would think of. They trust the system because they understand its limits. And they become the champions who carry it into their departments.

This is why training and building should happen together. A leadership programme that ends with slides leaves officers inspired but stuck. One that ends with a working prototype leaves them with something to test on Monday.

A safe way to start

For an agency that wants to move beyond the chatbot, I recommend a simple sequence.

1. Pick one process with clear rules, measurable delays and a willing owner.

2. Map the systems and data it touches, and decide what the agent may read, draft and do.

3. Build a prototype with the officers who run the process, using test data first.

4. Keep a human approval step on every consequential action, and log everything.

5. Measure time saved and error rates, then decide whether to scale.

None of this needs a national programme on day one. It needs one well-chosen process, a secure setup and a team willing to learn by building.

The future of public service is intelligent, automated and citizen-first. Agentic AI will not replace the judgement of public officers. Used well, it will give them back the time to use it.

Dr. Behrang (Hani) Parhizkar is Group CEO of Chamrun Digital and an agentic AI consultant and keynote speaker for governments and enterprises. Learn more at hanichamrun.com.