The commercial team needs a contract renewal clause interpreted before a meeting that afternoon. The query lands at eleven. It is not complicated, exactly: a standard restraint clause, familiar statutory backdrop. Getting it properly right means pulling the Act, checking the recent cases, verifying there is nothing in the last twelve months that has shifted how courts have read equivalent language. Done carefully, that is a morning's work. Done hastily, it is a risk.
Done by external counsel, it returns Thursday with an invoice that makes the CFO ask questions.
So she does not call. She has run the same calculation too many times. Forty minutes of proper research, billed at commercial rates, returns several thousand dollars for an answer she could form herself with the right foundation under her. For a novel question with material litigation risk, that cost is worth bearing. For a restraint clause on a standard commercial renewal, justifying it on a recurring basis is harder to explain to the CFO than the risk itself.
She tries a generic AI tool instead. The answer comes back fluent and authoritative-sounding. It cites a case she vaguely recognises. She cannot verify it. The tool has no source trail, no jurisdiction filter; it has predicted plausible legal prose from a training set that does not distinguish Australian contract law from its American or English cousins. Her professional obligation to verify sits with her regardless, and verifying a potentially hallucinated citation is not a workable foundation for advice. She closes the tab.
Manual research, then. She pulls the Act, opens the legal database, starts mapping the recent cases. It is thorough. It is also most of the morning. By the time she has a properly grounded answer, two more queries have arrived: a privacy notification question after a minor data incident, and a contractor classification issue a manager wants resolved before Friday. The restraint clause question is answered. The queue has not moved.
This is not a failure of process. It is the structural arithmetic of a lean in-house function. Each query is routine in isolation. Collectively, they consume the better part of a working week. The question every GC eventually faces is how to stay across the volume without building external-counsel reliance the budget cannot sustain.
Habeas's Research Assistants work differently from a search engine or a document uploader. You configure an assistant around a domain: contract interpretation, employment classification, privacy notifications, director duties. You give it a brief scope instruction. It retains that framing for every subsequent query in that domain, and when a question arrives, it conducts structured, agentic research and returns a cited analysis grounded in Australian primary law.
The restraint clause query that arrived at eleven goes to a Research Assistant configured for commercial contract interpretation. Within minutes, the output is there: the relevant Act provisions listed, the applicable case law mapped to the specific question, the interpretive framework stated with citations she can open and check against the source. The analysis is grounded in over 300,000 Australian cases and pieces of legislation, drawn from a closed dataset of legitimate Australian legal sources. The citations are real and linkable. The question is not whether she can trust the source trail; it is whether the analysis is complete and whether her professional read aligns with it.
The restraint clause question is answered before lunch. The privacy notification and the contractor classification question are in the queue. She puts both to their respective assistants.
Lisa, General Counsel at a Series A startup, described the experience directly: "It feels like having a law firm in your pocket. Not something you blindly bet the house on, but a powerful first-line legal intelligence tool."
That framing carries precisely the right weight. Research Assistants compress the analytical groundwork: identifying the right statute, mapping relevant case law, structuring the applicable test. They give the GC a solid foundation from which to form and communicate a legal view. The judgment remains with the practitioner. What Research Assistants remove is the scaffolding that had to be built from scratch every time.
The practical consequence for the relationship with external counsel follows from that. When first-line triage is handled in-house with proper grounding, external counsel can be reserved for what they are genuinely best placed to do: novel questions with no settled authority, matters with material litigation risk, specialist transactional work where professional accountability needs to sit with a firm. A GC who had integrated Habeas into her daily workflow put the trajectory directly: "External counsel used to be the thought partner. Now Habeas and AI fill that role. Over time, I expect legal spend on external counsel to be a fraction of what it is today, reserved mainly for highly specialised or strategic input."
Every routine query handled in-house, with traceable citations and genuine grounding in Australian primary law, is budget recaptured and capacity created. Foundational research processes that used to take a full morning can now be completed in minutes. For a GC fielding questions across employment, privacy, commercial contracts, and regulatory compliance in the same week, Research Assistants configured to those domains mean the queue shrinks without the analysis becoming shallow.
Habeas is direct about this. Particularly complex matters, questions where the law is genuinely unsettled, and anything with material downside risk still warrant careful professional judgment and, where appropriate, external engagement. The duty to verify remains with the practitioner. Verification on a Habeas output looks different from checking whether a general-purpose AI has fabricated an authority: the citations are real and traceable. The question is whether the analysis is complete and whether your professional read aligns with it. For the bulk of BAU queries flowing through a busy in-house function, that is a workable foundation.
If the eleven o'clock query sounds familiar, and if the answer still sometimes returns Thursday, there is a session available at habeas.ai. See whether a Research Assistant configured to your actual query domains changes the arithmetic.
The legal research in this article was conducted and every citation verified using Habeas, the Australian legal AI research platform.
Hero image: Invest Europe on Unsplash
