AI could help government solve its most stubborn technology problems

Somewhere in our public service, an essential government service is almost certainly running on software written before many of its users today were born. Throughout its lifetime, it will have been patched, upgraded and linked to systems its designers could scarcely have imagined. The people who first built it may have retired long ago.

Millions of Australians depend on these systems to receive payments, access healthcare and deal with government, so there is understandably little appetite to replace them without a carefully managed transition, given the essential services these systems support.

After three decades working with government, I get it: it is extremely hard to modernise vast systems while keeping critical services running. Medicare still has to process claims, Centrelink has to pay eligible claimants, and the tax office has to collect revenue, all while engineers figure out how all the IT fits together under the hood. Budgets are tight, and anyone who’s worked in government knows priorities can change quickly.

But something promising is happening today that I could not have imagined even a few years ago. Employed carefully and responsibly, artificial intelligence could finally make projects that have long seemed prohibitively expensive or risky considerably more achievable, to the benefit of all Australians.

Today's AI tools can examine vast volumes of code, identify dependencies between applications and help fill gaps in incomplete documentation. Since the start of 2027, I've seen technology projects traditionally expected to take three to five years that AI could help deliver in less than half the time, with smaller teams and at substantially lower cost.

The rate of improvement is almost dizzying, but this pace makes guardrails and checks more important than ever, particularly where payments, healthcare and personal information are concerned. Governments are right to be cautious: alarms were raised recently when an experimental OpenAI agent gained unauthorised access to a government Medicare statistics portal and retrieved internal files. No personal medical records were accessed, but the incident exposed the huge stakes if AI agents act beyond their instructions.

Finally, we have a technology that can reduce the herculean processing work involved in things like tax, immigration and social services. Behind every website and call centre are teams handling documents, checking information and moving cases between systems: AI could relieve much of this administrative load, allowing trained staff to concentrate on more complex, valuable work.

Publicis Sapient’s Digital Citizens Report 2025 found that more than half of Australians want government digital services to use generative AI, particularly to speed up approvals, reduce waiting times and make information easier to access. Since that report, more and more citizens are using AI to deal with government, whether agencies encourage it or not.

All of this means government information must be accurate, up to date and accessible to authorised AI systems. Until now, digital government has largely been designed around the assumption that people are accessing and navigating websites and menus themselves, but as AI assistants increasingly act on people’s behalf, they will need to draw on information from multiple departments, explain requirements and help users to complete applications.

This presents another opportunity for improvement, in what we tech specialists call ‘agent experience’: that is, how well government services work when a large language model acts on our behalf.

For historic and administrative reasons, our system of government is organised into departments and portfolios; each has its own responsibilities, funding and technology. But people’s lives rarely fit neatly into those buckets. Someone living with a disability may need healthcare, housing, transport and financial support. A person entering aged care may deal with several agencies and providers. The goal should be to make it easier for people to access support without needing to understand the structure of government.

Agencies such as Service NSW and Services Australia have made good progress toward making government easier to deal with. AI could, and should, take that much further.

Still, there is plenty of housekeeping to sort out first. Government information is scattered across databases, webpages, PDFs and documents all written at different times. Agencies will need secure ways for authorised AI systems to access reliable information and interact with different services. This raises challenging questions, such as who takes responsibility when an AI assistant gives incorrect advice about someone's entitlements or submits an application containing an error.

There is plenty of reason to progress with caution, but after three decades in this field, I am more excited than ever about the potential for technology to improve the way Australians interact with government. The greatest beneficiaries can and should be those who find government hardest to navigate, including people with disabilities, older Australians and those in regional and remote communities.

We should start with the technology government already has, using AI to understand complex legacy systems, identify opportunities for improvement and make them easier to modernise. After decades of incremental change, AI gives us a new opportunity to tackle some of the most longstanding challenges in government technology and ultimately make government simpler and more accessible for the people it serves. *Steven Metzmacher is Public Sector Lead at Publicis Sapient Australia.

Top image credit: iStock.com/Alexander Sikov