ITnews

How the Fair Work Commission safely deployed Agentic AI in weeks, not years


Prioritising strict governance and rapid development, Australia’s Fair Work Commission deployed governed AI agents into production in weeks, enhancing operational efficiency while maintaining public trust.



Australia’s Robodebt scandal left a scar on public trust in government automation. For government technology leaders, the message was unambiguous: automation without governance is a liability.

Joshua Pope, CTO of the Fair Work Commission (FWC), understood what AI could do for a tribunal that serves millions of Australians navigating workplace disputes. He also understood the stakes of getting it wrong. In a statement on operational reform, the Commission’s General Manager, Murray Furlong, pointed out that “by the end of financial year 2025–26, the Commission’s total workload will have increased by over 70% in the space of 3 years”, an operational strain that’s being compounded by the growing use of generative AI tools among the people the Commission serves.

Redefining what speed looks like in government, FWC moved from signed paperwork to onboarding with Workato within a single week, then immediately began building its first AI agent.

“Governance is really hard,” Josh says. “There was a very public bad experience with AI making choices on behalf of people, without a human in the loop. So for us, the ability to check, validate, and verify at every step was non-negotiable.”

FWC needed a platform that met the Australian government’s strict security and data sovereignty requirements, while giving its small technology team the agility to build AI solutions and push them into production at pace.

What separated FWC’s approach from the typical government procurement cycle was preparation. Before the contract was signed, Josh made platform training mandatory for his team.

“I specifically said I don’t want to see a single sales pitch as part of the onboarding week,” Josh says. “Just show me how to build stuff.”

The result was a procurement-to-production timeline that defies government norms. FWC moved from signed paperwork to onboarding within a single week, and attempted to build and deploy their first Workato AI agent on day one.

A custom IT Support Genie and License Optimisation Genie are now live internally, giving the Commission’s technology group a governed, repeatable process for deploying AI agents. Each solution sits within the compliance guardrails that post-Robodebt Australia demands, with explainability baked into every step.

While the AI Plan for the Australian Public Service 2025 sets an 18-month implementation horizon for every public servant to have access to generative AI tools supported by proper governance and training, FWC is already live.

FWC’s Infrastructure Manager, Kris Deep, leads the operational deployment of these agents, focusing on use cases that deliver value for his ICT team and the people they support. Their IT Support Genie is the centrepiece, customised with more than 30 skills.

FWC supports around 59 hearing rooms across the country, each running an average of 50 virtual and in-person hearings each day. A single room going down can delay a case someone has waited to have heard.

Status checks previously required Kris ‘s ICT team to log into four separate portals and manually review device statuses. A time-consuming, error-prone process that was only feasible once a day . If a hearing room went down mid-morning, the team often found out from frustrated users rather than from their own monitoring.

Now, with IT Support Genie, the team runs virtual conferencing checks multiple times daily across all systems in a single pass. The report flags devices with issues, affected rooms, polling health, unhealthy devices, and domain status.

“The Genie catches issues before our users do, and that changes the entire dynamic for my team,” Kris says.

The IT Support Genie handles much more : from provisioning overseas conditional access in five minutes (down from 30 minutes), to streamlining staff onboarding and offboarding (saving up to an hour per instance). All without needing the ICT team to intervene manually.

Built on Workato within a single week, FWC’s latest Bundle Document Genie agent pulls data from multiple sources to generate a master investigation report in two minutes, synthesising case notes often spanning more than 30 files. This saves at least an hour for associates who previously had to manually retrieve case information from a clunky web interface and Dynamics.

FWC’s approach to productivity carries a lesson that most AI deployments overlook. Freeing up capacity is only half the equation. The other half is knowing exactly where that capacity should go next.

“I’m very careful with giving time to people without a specific directive,” Josh says. “We’re not giving time back for people to decide what to do with it – they could fall into the trap of doing the same old things. We’re replacing that time with things we’ve already identified.”

The Commission is now extending its agentic capabilities to public-facing services, with a contact centre agent already in development. Murray’s statement highlights that the Commission’s helpline takes a high volume of calls on matters outside its jurisdiction, and confirmed it is assessing an AI voice agent to help triage those calls before they reach staff, with an aim “to reduce our call wait times and redirect queries as soon as possible”. Josh expected pushback from the frontline teams who would be affected by a change like that, but their reaction surprised him.

“Those at the frontlines said, no, triaging calls is the last thing they want to be caught up with. They wanted to focus on actually helping the people they connected with. So your better ROI comes from removing friction points in people’s day jobs, not from where you think you can see cost savings.”

Josh is a design partner for Workato’s Data Genie, providing real-world government input into the product’s development before general availability. His advice to other government technology leaders considering AI is practical.

“Find what you can reuse first,” Josh says. “Don’t get stuck researching where your internal ROI is and commissioning a bespoke solution every time. Research on where you can pick something up and just configure it for your use case.”

The Commission’s approach sits within a wider programme of reform, detailed in General Manager Murray Furlong’s May 2026 statement on change at the Commission.

For a closer look at why visibility alone does not amount to governance, see The Control Tower Fallacy whitepaper.



Source link