Artificial intelligence (AI) is evolving at a rapid pace. Emerging capabilities, like AI agents, offer new possibilities for NSW Government agencies.
We sat down with Daniel Roelink, Director Office for AI, Digital NSW, to discuss the role of AI agents and how agencies can use these systems safely and responsibly.
This article compliments the AI agent usage and deployment guidance. Check it out for more detailed information on using AI agents.
What is an AI agent?
An AI agent is software that can plan, decide, and then act to achieve a particular goal. ‘It’s quite advanced compared to what we are used to seeing with simple AI chatbots in that it has a greater level of autonomy,’ says Daniel. ‘AI agents don’t need to be given instructions; the agent itself can work out what it needs to do and then execute.’
While simple AI chatbots produce responses, AI agents learn, use tools, remember context, and work with other agents to achieve goals. ‘They’re quite powerful, and with that, there’s a few things you need to be mindful of,’ adds Daniel.
New guidance for agencies
To help NSW Government agencies make informed decisions about using AI agents, Digital NSW have published AI agent usage and deployment guidance.
'It’s important we publish guidance for emerging AI capabilities to help agencies define their policy position, ensuring they consider new risks and opportunities while using this technology to benefit the public service and improve the services we provide.'
- – Daniel Roelink, Director Office for AI, Digital NSW
Developed alongside industry chief AI architects, the guidance outlines key considerations to assist agencies in developing policies and assurance processes. It also includes checklists to help determine if an AI agent is needed and what steps should be taken before piloting and deploying one.
‘There are many questions you need to consider when using a new technology,’ says Daniel. ‘The value of this guidance is that it sets out a consistent set of questions to be considered, ensuring proper evaluation and oversight of AI agent use.’
Test in controlled environments
For agencies considering using AI agents, Daniel recommends assessing the agent first in an internal, low-risk environment. ‘Don’t let the agent decide what it can do or which systems it can access,’ says Daniel. ‘Be strict about its permission from the start.’
Restricting the agent to read-only access and limiting the number of platforms it can interact with is also important, especially in the early stages of deployment. ‘The more you integrate AI agents into other systems, the harder it is for you to have governance and oversight of the data,’ explains Daniel.
Daniel also advises to always start with a single agent, as new risks are introduced in multi-agent deployments. ‘When multiple agents communicate, misinformation can be amplified and quickly spread,’ says Daniel. ‘Make sure you start with one agent and exercise caution before adding more.’
Establish clear ownership
Before deploying an AI agent, it’s important that agencies assign a business owner to that agent. This ensures clear accountability and responsibility. ‘If you’re deploying an agent, it has to be tied to an organisational role so that any issues can be managed,’ says Daniel.
‘You also have to make sure you have visibility of that ownership centrally, so if the organisation changes or people leave, you can then transfer that accountability,’ adds Daniel.
Opportunities for government
As an emerging capability, AI agents present promising opportunities for NSW Government agencies.
‘AI Agents are unique because they don’t just respond – they can act on what they know, interact with other systems, and guide what happens next.’
- – Daniel Roelink, Digital NSW
AI agents can be particularly useful for low-risk, high-volume use cases. For example, an AI chatbot being used for IT service requests could trigger an AI agent to initiate a workflow, such as logging an incident or creating a service request.
As agencies explore these opportunities, Daniel advises applying the same best practices used in previous technology implementations. ‘Follow your learned experience in best practice of digital, and apply that thinking,’ says Daniel. ‘AI introduces some difference, which is why we’ve published these guidelines, to highlight them and help educate people.’
What’s next?
AI agents are still in the early stages of adoption, and Digital NSW will continue to expand and refine its guidance as global standards evolve. ‘The next step for us is to integrate AI agent considerations into our AI Assessment Framework,’ says Daniel. ‘We will continue to provide general guidance, while also offering tailored advice based on real-world use cases.’
Before using AI agents, agencies should consult their governance and assurance teams to identify all relevant requirements for their business area.
To learn more about using AI agents, explore the AI agent usage and deployment guidance.