Codify first, then automate: what five HR leaders said about GenAI in practice

The Harnessing Generative AI panel at HR Tech Fest in Sydney, facilitated by Paul Swain of Ultimate People, put people leaders from Atlassian, CBRE, Anglicare Sydney, Loan Market Group and Latitude Financial on one stage. They spanned organisations of 450 to 160,000 people. The room rated itself mostly "operational" on GenAI maturity, with very few experts. The panel's advice suited that room: start narrow and be specific.

AI can't use what it can't see

The first piece of advice was to codify knowledge before anything else. If your policies, processes and judgement calls live in people's heads or in an eight-page PDF nobody reads, no agent can help. The panel recommended picking three to five high-volume, high-friction use cases instead of spawning hundreds of agents. Give each one a single accountable owner. Treat clean, current data as a prerequisite.

The use cases they described were practical:

• Policy interpretation (Loan Market Group). Staff ask a question in plain language and get a tailored answer instead of a PDF.
• Goal checking (CBRE and LMG). An agent tests objectives against SMART criteria or OKR structure. CBRE found around 40% of goals previously failed that test.
• Onboarding and pay conversations (Atlassian). NORA answers most new-hire questions. COCO helps more than 2,000 managers deliver explainable merit and promotion outcomes.
• Emergency capacity (Anglicare). Franco, a fuel hardship agent, was built in an afternoon to process a flood of applications when there were no staff to do it.
• First-pass review conversations (Latitude Financial). AI guides leaders through performance conversations, with a human fallback kept in place.

Why this matters for smaller HR teams

Every example above started with an organisation writing down how it does something, then automating the repeatable part. Large organisations have the people to do that codifying. A solo HR practitioner or a 40-person business doesn't.

That is most of the work behind Nooma, the AI practice companion Castle designed and built with O-HR. The workflows and audits in the platform are HR practice written down, step by step, so the AI has something solid to work from. The practitioner doesn't have to codify their own offboarding process before AI can help with it. See how we built it.

Measure more than time saved

CBRE measures three things: financial value (time saved multiplied by volume), experience, and the share of work an agent handles. Atlassian tracks adoption, efficiency, decision quality with a human in the loop, and whether roles are actually being redesigned. It also sets super-user thresholds per function, not company-wide. Anglicare tracks containment, meaning how much its contact centre chatbot resolves without a handoff.

One detail surprised the room: at Anglicare, registered nurses are the top Copilot users, not back-office staff.

The personal version

Atlassian's framework for individuals was the most useful takeaway for practitioners:

• Avoid: stop doing what AI can do.
• Automate: remove repetitive tasks entirely.
• Augment: use AI as a sparring partner.
• Elevate: reclaim the high-value work the grunt work was crowding out.

The panel's closing point was that tech capability is now non-negotiable for People and Culture. CBRE's Chief People Officer is among the top AI users in a 160,000-person organisation. At Anglicare, the CHRO now also oversees IT, brand and shared services. Latitude Financial's Chris Best built and deployed an app himself with no coding background. When HR leads, the business reads it as a sign that broad use is safe.

If you're building AI into specialist work of your own, here's how we approach it. For HR practitioners, Nooma launches to founding members soon: join the waitlist.