The question practitioners are increasingly asking an AI legal research tool is something like: "Is this restraint of trade clause enforceable?" That is an understandable question. It is also the wrong one, and how the profession frames it will shape whether AI makes restraint of trade advice sharper or quietly worse.
Restraint of trade doctrine in Australia is, by design, hostile to certainty. The test from Buckley v Tutty (1971) 125 CLR 353 requires reasonableness in all the circumstances, as between the parties and from the public interest perspective. That formulation is not a gap in the law waiting to be filled by better cases. It is the doctrine's architecture. Courts are being asked to weigh, proportionally, a set of facts that will never be identical twice.
Recent decisions reinforce how granular that weighing gets. The cases practitioners are watching on post-employment restraints and distribution agreements turn on things like the precise scope of customer relationships the employee actually held, whether the geographic boundary maps onto the territory where competitive harm is genuinely plausible, and whether a cascading duration clause was realistic or cosmetic. Small factual shifts produce different outcomes. The doctrine tells the court what to do. The facts tell the court what to decide.
This is where the distinction between what AI does well and where it hits a structural limit becomes practically important.
Where AI earns its place
Case law in this space is voluminous and not consistently indexed by legal concept. Gathering the factual patterns courts have accepted as legitimate business interests, tracing how duration and geographic scope have been treated across employment versus commercial contexts, identifying which factual vectors have been outcome-determinative in comparable configurations: all of this is extraction and synthesis work that used to take hours and now takes minutes. AI does it accurately if the underlying sources are authoritative Australian primary law.
The same applies to principle mapping. When a client hands over a cascading restraint clause and asks whether courts have enforced similar constructions, AI can rapidly surface the doctrinal landscape, flag tensions between different factual clusters, and identify which competing principles are likely to govern. That is not a trivial contribution. It is the scaffolding that lets a lawyer spend their time on the judgment call rather than the retrieval.
Where it stops being analysis and becomes something else
Proportionality is a judicial exercise. The question of whether a twelve-month restraint across a defined metropolitan area for a senior account manager in a specialised industry is reasonable is not a research question. It is an evaluative judgment that depends on facts not yet fully developed, a client's risk appetite, the likely forum, and the practitioner's read on how a particular court or judge weighs competing equities.
AI cannot tell you whether the legitimate business interest the employer claims is sufficiently cogent to justify the scope. It cannot assess the credibility of the employee's claims about their actual role. It cannot predict how a court will apply proportionality to facts it has not yet seen. Any tool that appears to give a confident enforceability verdict on a set of instructions is not doing legal analysis. It is pattern-matching with a confidence problem.
Why this matters now
The risk is not that practitioners will use AI instead of thinking. It is more subtle: that AI-assisted research creates an impression of thoroughness that substitutes for the harder exercise. A well-structured case cluster summary can feel like an answer when it is actually organised input to the answer still requiring your judgment.
Restraint of trade is a useful test case for AI legal research precisely because its fact-sensitivity is non-negotiable. The doctrine cannot be resolved by better retrieval. But better retrieval, faster and more accurately anchored in Australian primary sources, changes what the lawyer does with the time between instructions and advice.
That is the appropriate ambition: AI sharpens the inputs; the judgment remains yours.
Habeas is built for the extraction and synthesis part of that workflow, with research anchored in Australian primary sources and cited to paragraph level, so practitioners can interrogate the principles before they apply them.
