You made hundreds of decisions today before you finished your coffee. What to wear. Which email to answer first. Whether to reply to that message or leave it. None of them needed a hearing.
Now think about a different kind of decision. A tribunal member deciding whether a visa applicant stays or goes. A judge deciding whether someone is guilty. A claims manager deciding whether a policy responds. We reach for the same word, decision, for the coffee and for the courtroom. For anyone building software in this space, that looseness is worth stopping on, because the two things are not the same, and the difference is where the interesting questions live.
So it's worth asking plainly: what actually is a decision, in the places where it changes someone's life?
The part everyone sees, and the part they don't
A decision in dispute resolution has features your morning choices don't. It's contested: two accounts of the world, competing. It's authorised: the person deciding holds delegated power to do it. It's reasoned: the why is part of the decision, not an optional extra. And it's binding: orders flow from it, and the world rearranges itself accordingly.
But here's the thing that a career in and around courts teaches you: the moment of decision is the smallest part. It's the tip of an iceberg. Underneath sits the real bulk of the work: gathering the evidence, marshalling the documents, framing the issues, working out which rules apply and to what. By the time a decision-maker sits down to actually decide, most of the labour is already done. What they're deciding on has been shaped by a hundred smaller choices along the way: what went in the bundle, what got summarised, what was called relevant, what weight the evidence carried.
Some of those smaller choices are decisions in their own right. Others are just… steps. And reasonable people disagree about which is which. That ambiguity isn't a footnote. It's the whole question.
This shape is older than the common law
Look back far enough and the pattern holds across systems that had nothing to do with each other.
In classical Athens, a jury of 201, 501, sometimes more than a thousand citizens heard the speeches, voted by ballot, and decided the same day. No judge and no reserved reasons, but also no apparatus: the parties did their own preparation, and the "decision" was a raw count of votes.
In imperial China, the county magistrate was investigator, prosecutor and judge in one person, and he could not function without the twenty or thirty clerks, runners and secretaries of the yamen behind him, preparing everything he ruled on.
In England, Henry II sent his judges out to ride circuit, giving consistent reasons town by town and slowly building the common law on the road. But the writs, the pleadings and the local juries were prepared and waiting before the judge ever arrived.
Three systems, two and a half thousand years apart. In each, the human at the apex did the deciding, and in each, a large machine underneath did the preparing. The decision was always the visible peak of an invisible mountain of administrative work. That hasn't changed. What's changed is that, for the first time, a lot of that administrative work can be done by software.
Software has been resolving disputes for twenty years
This is not speculative. When Colin Rule built eBay and PayPal's dispute resolution system in the 2000s, it grew to handle more than 60 million disputes a year, more than the entire US civil court system. Over 90% were resolved without any human third party deciding at all. Rule's line has stuck with me: at eBay, people would rather lose a dispute quickly than win one slowly, and "slowly" meant twelve days.
So the question for courts and tribunals was never can software resolve disputes at scale. It plainly can. The question is which disputes, at what stakes, and under whose authority. An eBay refund and a protection visa are not the same animal. On eBay, both parties agreed to the process by ticking a box, the stakes are a few dollars, and legitimacy comes from a contract. In a tribunal, no one ticked a box, the stakes are someone's liberty or livelihood, and legitimacy comes from the public. You cannot lift the mechanism from one and drop it into the other without asking what you're carrying across.
We are drowning, which is exactly when shortcuts get tempting
The pressure to carry too much across is not abstract. It's the backlog.
In England and Wales, outstanding Crown Court cases passed 76,957 at the end of March 2025, with trials now being listed as far ahead as 2030. In India, the courts were carrying roughly 5.39 crore (about 53.9 million) pending cases at the end of 2025. Closer to home, Australia's Administrative Review Tribunal was sitting on around 124,000 matters in late 2025; of the cases it managed to finalise across one recent stretch, 44% came back with the original decision changed.
Sit with that last one. Nearly half of a national tribunal's finalised matters overturned or varied the decision below. These are not abstract numbers. Each is a person waiting for a visa, a verdict or a payout while the queue grows. Backlogs on this scale aren't primarily a resourcing problem. They're a design problem. And when a system is drowning, "let the software just decide it" starts to sound like mercy rather than the risk it actually is.
What the law means by "decision", and why AI complicates it
Here's where it gets genuinely hard. Australian law has wrestled with the meaning of "decision" for decades. In Australian Broadcasting Tribunal v Bond, the High Court held that a reviewable decision is generally one that is "final or operative and determinative": a substantive conclusion, not every step on the way to it. A finding en route to the outcome usually isn't itself a "decision."
That doctrine was written before a machine could perform the steps. So put the modern question to it directly: when an AI summarises a witness statement, or classifies a document as relevant, or ranks the issues for a decision-maker: is that a "step on the way"? Or has it quietly done something more, by shaping what the human sees before the human decides anything?
I don't think there's one clean answer, and I don't think there should be. It depends on the institution and the matter. In the High Court, the exercise of judicial discretion is the whole point, and no one sane wants to automate it. In a parking fine, the discretion is thin and the stakes are low. On appeal from that same parking fine, discretion matters again. The line between "administrative task" and "the decision itself" moves depending on where you're standing, and pretending otherwise, in either direction, is how you get it wrong.
The courts themselves are already drawing the line in practice. The NSW Supreme Court's practice note on generative AI lets practitioners use it for chronologies, indexes and summarising documents (the preparation), while the accompanying guidance for judges is explicit that AI is not to be used to assess evidence or write the reasons for judgment. Assist the runway; stay out of the cockpit.
Justice Lee put the principle about as well as it can be put, in a 2026 judgment about company directors leaning on AI to digest their board papers: "The use of technology may assist comprehension, but it cannot displace judgment." He was talking about directors, not judges. That's the point: the same boundary shows up wherever accountability is personal. It's true in a boardroom, a courtroom, a tribunal and a claims team alike. Comprehension can be assisted. Judgment cannot be handed off.
Where Briefcase plants its flag
This is the line we build to. Briefcase is case management software that uses AI to shorten the time to a human decision, not to make the decision. We work on the mountain, not the peak: the evidence, the bundles, the summaries, the chronologies, the administrative weight that buries decision-makers and delays everyone waiting on them. The act of deciding stays with the person the public has entrusted to make it.
And I want to be honest about the harder question underneath, because it doesn't belong to us to answer. Where the line between assistance and decision sits, and how much of the run-up to a decision we're comfortable letting a machine touch, is not a technical question, and it is not one that the decision-making institutions, or their software vendors, should settle quietly on everyone else's behalf. It's a public question. It's the kind of thing a society works out in the open, with judges, regulators, lawyers, technologists and the people whose lives these decisions land on all in the room.
We have a strong view about where that line sits today. We'll keep arguing for it. But the deciding about the deciding belongs to more people than us.
So: what is a decision? It's the moment, but it's mostly the mountain. Software can carry the mountain. Who carries the moment, and how much of the climb we let a machine do before we get there: that's the conversation worth having, and it's only just started.
Briefcase is an AI-native case management platform for courts, tribunals, regulators and dispute resolution bodies. We build software that speeds up the administrative work around a decision so the people responsible for making it can focus on the decision itself.