Three large air tankers and their lead plane were sent to the Tenterfield area on 31 October 2023, close to the New South Wales border with Queensland. Up to 21 Rural Fire Service aircraft were working the firegrounds there that afternoon, those four among them, and Queensland was flying its own on the other side of the line. The tankers arrived without a target. The air attack supervisor coordinating aircraft overhead did not know they were inbound. Several times that day, crews came closer to other aircraft than was safe (Australian Transport Safety Bureau, 2025).

The Australian Transport Safety Bureau investigated. What it found had less to do with individual error than with a missing instruction. The NSW RFS had no procedure requiring the state air desk to coordinate a large air tanker tasking with the incident management team on the ground. Neither the NSW RFS nor its Queensland counterpart had established cross-border procedures for aerial firefighting. Everybody involved was competent, and everybody was working inside the same incident management system. What was missing was a written answer to one question: who connects a state-level decision about a national asset to a local incident plan.

I have been re-reading that report this week, because of two documents published two days apart.

On 19 August, AFAC released the Seasonal Bushfire Outlook for Spring 2026 at the AFAC26 conference in Melbourne, with the Bureau of Meteorology among the nine agencies that developed it. It identifies an increased risk of fire, which AFAC’s map legend defines as “the likelihood of an increased number of significant bushfires occurring in the outlook period compared to the average risk of fire”. That category runs from northern and south-eastern New South Wales through to large parts of the Northern Territory. The ACT is the only jurisdiction it leaves out (AFAC, 2026b). The conference theme was “Leading together: Integrity, inclusion and impact”. In the Bureau’s release, AFAC President Darren Klemm, who is also Western Australia’s DFES Commissioner, described the Outlook as “a great example of our agencies working together to share knowledge and provide communities with a clear picture of what the upcoming season is going to be like” (Bureau of Meteorology, 2026).

Two days later, Communications Earth & Environment, a Nature Portfolio journal, published a Perspective on the January 2025 California fires. Retrospective analyses of those fires reach for climate and fuel, the authors argue, while missing “a critical systemic blind spot”: the inefficiencies in suppression that come from misaligned incentives and fragmented interagency coordination (Debnath et al., 2026). The proposed remedy is agentic AI operating inside a governance framework that aligns incentives and assigns accountability.

Two things should be said about that paper before it is used for anything. It contains no original data. Its authors describe it as “a conceptual commentary … rather than a deployment ready operational plan”, resting on the Los Angeles City Fire Department’s after-action review of the Palisades fire, a McChrystal Group report on alert and evacuation systems, and the game-theory literature on cooperation. It is also entirely about California. Applying its diagnosis to Australia is a move the reader has to make for themselves, because the paper never makes it.

The diagnosis still deserves to be taken seriously, because Tenterfield is the same shape. The remedy is what deserves an argument.

That remedy, for a multi-agency coordination failure, is a multi-agent AI system. It is a serious proposal and the governance half of it is the more interesting half. But hold it against the fireground above. At Tenterfield, human agencies working to one doctrine, with statutory accountability behind them, could not connect a single tasking decision to a single incident plan, because no procedure said who owned that connection. Software agents do not supply the missing procedure. They add participants to a decision that already had no owner, and they make the ownership question harder to answer afterwards, because the record of who decided what now includes parties that cannot be called to an inquiry. The objection here concerns sequence. Incentive and accountability design has to exist before the automation arrives, or the automation inherits the ambiguity and scales it.

The deeper trouble is that “coordination” flattens several problems that have different owners and different fixes. In the Australian context, at least four such problems sit under that one word: sharing resources across state borders; who holds authority to warn at the interface; tasking aviation into a local incident plan; and where agency responsibility ends and land manager responsibility begins on fuel. Treating them as a single cultural failing produces a single cultural remedy, which is usually a communications programme. Communications was never the constraint.

Separate them and the picture improves considerably. Cross-border resource sharing is the one Australia has largely solved, through a standing mechanism rather than goodwill. After Victoria’s catastrophic fire conditions on 9 January 2026, the AFAC National Resource Sharing Centre moved 500 resources into Victoria over six days, along with 47 tankers from New South Wales and South Australia. At the time of reporting, 342 interstate personnel from six states and territories were still deployed, alongside national aviation assets rotating between them (AFAC, 2026a). Aviation tasking into a local plan is the one that was unsolved as recently as 2023, and it has since been addressed: the RFS now requires large air tanker pilots to be briefed on incident strategy before departure, and a national cross-border airspace guideline followed.

The evidence from comparable jurisdictions suggests the answer is not simply more centralisation or less. Park and Nam (2026) compared the 2025 interface fires in Japan and South Korea, two countries with similar terrain and opposite organisational models. Japan runs a decentralised municipal response with volunteer brigades; South Korea runs centralised command with better aerial assets. Both failed at the interface. In the Sancheong fire, municipal crews were on the ground within 17 minutes while specialist national units took up to 108, and the resulting mismatch delayed the establishment of unified command. Structure alone did not decide the outcome.

The United Kingdom’s experience points the same way. A University of Liverpool study of the Joint Emergency Services Interoperability Programme found that embedding it was hindered by the perception that it was “focused on blue light responders”. More tellingly, the study also found that pre-existing organisational policies and legislation sometimes conflicted with the doctrine itself (Waring et al., 2025). More than a decade of shared interoperability doctrine does not override an agency’s own statutory obligations. In Australia the equivalent gap sits around the parties outside the fire-service core, and land managers are the clearest case of it.

None of this argues that Australian coordination is failing. AIIMS already gives this country the shared incident management system JESIP is still trying to embed, which is precisely what makes the failures that do occur interesting: the shared doctrine was already in place when they happened. The same fortnight produced real coordination as well. CFA, with an Australia-wide AFAC working group, launched Australia’s first nationally consistent bushfire safety training for workers, a four-topic module for employers and their staff in sectors running from aged care to agriculture. Nothing national had existed for that audience before (CFA, 2026). Natural Hazards Research Australia signed a memorandum of understanding with New Zealand’s Natural Hazards and Resilience Platform at the same conference (Natural Hazards Research Australia, 2026).

There is also a live Australian test already on the table. Victoria’s parliamentary inquiry into the 2026 summer fires tabled its final report on 28 July with 93 findings and 80 recommendations, spanning fuel management through to recovery coordination (Environment and Planning Committee, 2026). The Government has six months to respond, which places its answer around late January 2027, after the season the Outlook is describing has begun.

One small illustration of how easily this slips. Count the jurisdictions in AFAC’s Outlook and you get seven, because AFAC includes South Australia. The Bureau published its own release about that same Outlook the next day. Count the jurisdictions there and you get six, because the Bureau leaves South Australia out. Nobody in either organisation was careless. Two organisations simply described one document slightly differently, and anybody quoting a jurisdiction count this week could have quoted either number in good faith. That is coordination at its most benign, and it runs on the same mechanism that put three tankers over a fireground where nobody was expecting them.

Warning the public is a separate problem, and I have written about that one elsewhere. This is about the agencies reaching each other. The Outlook tells every agency how hard the season is likely to be. It does not tell them which of those four decisions currently has a name against it. Somebody in each agency could write that list down before the season starts.

References