You've Just Been Made Chief AI Officer. Now What?

By July 31 2026, every Australian federal agency is required to have appointed a Chief AI Officer. The mandate comes from the APS AI Plan 2025, released in November last year, and it is not optional. What is optional, apparently, is a clear playbook for what those people are actually supposed to do once they have the title.

Most CAIO appointments are going to existing senior executives. SES level 1 or above, absorbed into a current role, no new headcount, no dedicated budget. The person who was already the CIO, the CDO, or the Deputy Secretary of Corporate Services is now also the Chief AI Officer. They have the accountability. They have the mandate. What they often don't have is a clear picture of where to start or how to make the role mean something beyond the compliance checkbox it was created to satisfy.

This post is an attempt to provide that picture.

What the Role Is Actually For

The APS distinguishes between two positions that often get conflated. The AI Accountable Official is responsible for governance: ensuring the agency uses AI responsibly, manages risk, and complies with policy. The Chief AI Officer is responsible for transformation: identifying where AI can improve outcomes, driving cultural change, and making adoption actually happen.

The distinction matters because the failure mode of a CAIO who defaults entirely to governance is an agency that is very carefully not doing much. Risk management is necessary but it is not the job. The job is to move the agency forward on AI capability in a way that produces measurable improvement in how it delivers for citizens, while maintaining the trust that makes that delivery legitimate.

In 2026, that is a genuinely hard problem. Not because the technology is unclear, it is more capable and more accessible than it has ever been, but because the organisational, cultural, and governance conditions that make AI adoption successful in a public sector context are different from what private sector case studies describe, and the gap between mandate and capability in most agencies is wider than anyone is saying publicly.

The Three Problems Nobody Briefed You On

The data problem is worse than you think. Every AI application of any consequence is only as good as the data it can access or be grounded in. Government agencies, almost without exception, have data quality, consistency, and accessibility problems that have been accumulating for years. Legacy systems that don't talk to each other. Information that exists in one division and is invisible to another. Records that were captured in formats that made sense in 2008 and create significant friction now.

None of this is a reason not to proceed with AI adoption. It is a reason to be honest about the sequencing. The agencies that will show the best outcomes in twelve months are the ones that invested six months in understanding and improving their data foundations before building AI applications on top of them. The agencies that skipped that step will have a collection of pilots that don't perform as expected and a growing credibility problem with their staff and their ministers.

The skills gap is structural, not individual. The APS AI Plan mandates foundational AI literacy training for all 220,000-plus APS employees. That training is important and genuinely needed. It is also not sufficient preparation for the people who are going to be designing, procuring, and overseeing AI systems. There is a meaningful difference between understanding what AI is and being equipped to evaluate whether a vendor's proposal is credible, specify a system's requirements accurately, or oversee its operation once deployed.

The CAIO who treats skills development as a training completion metric rather than a capability building programme is going to find, within twelve months, that the organisation has a lot of people who completed the module and not many people who can do the work.

The governance framework has to be real, not cosmetic. The responsible AI policy requirements are genuine obligations, not paperwork to be satisfied and filed. In a public sector context, the stakes of an AI system producing a wrong or biased output are different from the private sector equivalent. The citizen who was assessed incorrectly by an automated system, the decision that was influenced by a model that wasn't operating within its validated parameters, these are not incidents with a PR cost. They are failures of public trust that erode the legitimacy of the whole programme.

A governance framework that exists as a document rather than as a living operating procedure is not a governance framework. The CAIO who builds oversight that actually catches problems, who creates feedback mechanisms that surface failures, and who is prepared to pause or withdraw AI systems when they're not performing to standard, is the one who earns the trust to keep building.

A Practical Day-One Framework

If you have just been handed this role and are working out where to start, here is the sequence that produces the most momentum with the least waste.

Assess before you promise. Before committing to any specific AI initiative, spend four to six weeks on an honest capability and readiness assessment. What data does the agency have and in what condition? What processes are well-documented enough to automate? What skills exist and where are the gaps? What AI tools are already in use, formally or informally, and what does that tell you about where demand and capability already exists? This assessment is not a delay tactic. It is the difference between a programme built on accurate foundations and one built on assumptions that will break later.

Find the quick wins that are actually quick. Most agencies have a set of high-volume, well-defined, low-risk processes where AI can reduce administrative burden quickly and visibly. Document summarisation. Meeting transcription and action extraction. First-draft policy research. These are not transformational. They are the proof points that build confidence, develop capability, and generate the credibility to pursue more ambitious applications. Start here, not with the application that would look best in a minister's briefing.

Build the governance structure alongside the pilots, not after them. The instinct is often to get something working and then think about governance. This is exactly backwards in a government context. Every pilot should have a defined human review process, clear criteria for what triggers escalation, and documented accountability for the output. This adds time to early pilots. It saves considerably more time when something goes wrong in a later one.

Get your peers around the same table. The AIDE function exists specifically to coordinate CAIOs across agencies. Use it. The problems you are encountering are being encountered in parallel by dozens of other agencies, and the solutions being developed in one context are often transferable to another. The CAIO who tries to solve everything independently is leaving significant value on the table.

The Strategic Framing That Changes the Conversation

The CAIO role, done well, is not primarily a technology role. It is a change leadership role that happens to be triggered by technology.

The hard work is not procuring the right tools or deploying the right models. The hard work is helping an organisation that was built around a particular way of working understand why and how that needs to change, while maintaining the trust and accountability that public sector service demands. That requires the ability to communicate honestly with both technical teams and ministerial offices, and to make sequencing decisions that are right for the long term even when short-term pressure is pointing in a different direction.

The agencies that navigate this transition well will be the ones where the CAIO saw the role that way from the start. The ones that don't will have delivered a lot of pilots and produced little lasting change.

The deadline is weeks away. The work is just beginning.

If you're stepping into a CAIO role or supporting an agency's AI transition and want to think through the practical framework, the governance structure, or the sequencing decisions, I work with public sector and enterprise leaders on exactly these problems. Happy to have a direct conversation.