The Verb Spellbook Shouldn't Have Used

Spellbook claims to automate contracts end-to-end. But does autonomous contract management deliver on its promise for Australian lawyers?
Green cactus plant on a white table, symbolising resilience and caution in evaluating AI contract management claims.

The email arrives at 11 pm. Or it comes through Slack, or it's embedded in a Salesforce record, or it's attached to a Teams thread that nobody thought to flag. It is a contract, and someone needs to do something with it. By morning, there are four more.

This is the inbox problem that contract management has tried to solve for years, and it is largely unsolved. Spellbook recently launched its Autonomous Contract Management product, billing it as the first AI system to run contracts end-to-end: intake, redline, negotiate, renew, all before a lawyer opens the document. The announcement landed quickly in legal tech coverage. We have been watching Spellbook for a while. They build good tools. ACM has parts worth paying attention to, and one claim worth reading with both eyes open.

What Spellbook Built

ACM pulls contracts from email, Slack, and Salesforce, compares them against a team's playbook, and redlines to that standard. The system handles renewals and tracks obligations. Lawyers can configure the playbook parameters, and the tool applies them at scale and speed no human can match across a high-volume contracting workflow.

For an in-house team drowning in NDAs and SaaS agreements, the throughput case is real. Standardised, low-stakes agreements reviewed against a defined playbook is the work that benefits most from automation. If the parameters are well-set and the document types are familiar, ACM does something useful: it handles the queue.

The part worth pausing on is the framing: "before any lawyer opens them."

The Autonomy Claim

Spellbook is marketing ACM as autonomous. The word is doing significant work in that press release. An AI agent that handles intake, redlines, negotiates, and renews a contract without a lawyer in the loop is, in our view, a professional-responsibility description sold as a product feature.

Someone owns every clause in every agreement that goes out under a company's name. In Australia, that person is the GC, the company secretary, or an authorised officer. The fact that no lawyer opened a document before it was signed does not create a liability shield. It creates a liability gap.

Consider a NDA where ACM redlines the residuals clause to the playbook standard and the counterparty accepts. The company later discovers the clause permits the counterparty to retain information in the unaided memories of employees who accessed confidential materials during the negotiation. A lawyer reviewing the redline before execution would likely have caught that. Nobody reviewed it. The agreement is signed. The question at that point concerns who is accountable for that mistake making it into an executed agreement. In Australia, the answer is the same as it has always been: the officer who allowed the company to be bound.

We are not saying ACM cannot be used responsibly. The question for any GC considering it is: what is the review trigger? At what point does a human see this before it binds the company? If the answer is "after execution," that is where the exposure sits.

"Autonomous" is a marketing verb. It describes what the tool does without the human. It says nothing about what happens when the autonomous output turns out to be wrong.

The Tell in the Small Print

Here is what we found more interesting than the headline claim. Buried in the ACM feature set is a searchable repository of signed agreements, with citations traceable to the source documents themselves. The "autonomous" intake piece gets the press release. The cited, searchable repository is what a GC lives in day to day.

That feature exists because in-house teams have learned through experience: a contract management system earns its keep in two ways: processing new agreements and surfacing what the company has already committed to. A GC asked about a specific indemnity cap, a warranty position, or a change-of-control clause needs to locate the relevant agreement, confirm the language, and be able to show her work. Spellbook added that repository because traceability is the thing in-house teams pay for, whether or not the product language foregrounds it.

The autonomy story goes on the press release. The citation store is what survives contact with a commercial dispute.

For Australian GCs

The Australian context adds a layer the announcement does not address. ACM is a Canadian product built for North American contracting workflows. The playbook parameters, the standard positions, the concept of what a "market" clause looks like: these are calibrated to a different jurisdiction.

The unfair contract terms regime under the Australian Consumer Law is one concrete example. For standard form contracts with small businesses or consumers, terms that create a significant imbalance in the parties' rights and obligations can be declared void under the Competition and Consumer Act 2010 even if both parties signed. An AI playbook built around North American standards would not know to flag that. Similarly, the Fair Work Act 2009 sets minimum conditions that cannot be contracted out of. A playbook that accepts employment contract terms unremarkable by US standards may produce agreements that are unenforceable in the provisions that matter most to the company once a dispute arises.

For a GC managing Australian employment contracts, Australian consumer law obligations, or agreements governed by the Competition and Consumer Act, the playbook problem is material. A redline reflecting North American standards on limitation of liability may not reflect what an Australian court would consider reasonable. The speed gain is real, and the jurisdiction gap is also real; it matters most when the output gets relied on.

AI-assisted contract work remains worth pursuing. The discipline is being precise about what the tool knows, and where its knowledge ends.

Anti-Hype as a Design Principle

We have said for a while that honest AI in legal work looks less like autonomy and more like augmentation. An agent that surfaces the relevant clause, flags the deviation from the playbook, and presents the issue to a lawyer in a form the lawyer can verify is more useful than an agent that resolves the issue before the lawyer is involved, because the lawyer's judgment is the part that carries the professional and legal weight.

Habeas was built on a different premise. Research Assistants built on Habeas return structured, cited analysis: authorities mapped to each issue, grounded in Australian primary law, with every source traceable to the document it came from. The GC who uses Habeas to assess whether a particular restraint-of-trade clause would be enforceable gets an answer she can check. The cases cited are real. The statutory provisions are correct. She can disagree with the analysis and point to what she disagrees with, because the analysis is auditable. That is what AI for lawyers looks like when it is designed around the practitioner's professional responsibility.

One GC we work with described it as "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 matters: the tool informs judgment rather than replacing it. The corpus behind that research covers over 300,000 Australian cases and pieces of legislation, scanned in seconds, from a closed dataset of verified Australian legal sources. Research that used to take a full morning can now be completed in minutes. The speed is real, and the traceability is what makes the speed usable.

Where to Land on ACM

Spellbook's ACM is worth watching. High-volume, standardised contracting workflows are a genuine problem, and a well-configured playbook applied at scale is a partial solution worth having. The repository and citation features are the parts we would build a workflow around.

The autonomy claim is where scepticism earns its keep. Any Australian GC who adopts ACM will want to be clear, before it goes live, on the review architecture: which documents trigger human review, at what stage, and who is accountable for sign-off. That architecture determines whether the tool is being used responsibly or whether the speed gain is being purchased at the cost of the oversight layer that professional duty requires. "The AI reviewed it" has never been a defence to a contractual dispute.

The broader signal from the Spellbook launch is the same one we have been reading in every "AI is coming for legal work" announcement for the past two years. The tools are getting better, faster, and more capable. The professional judgment of the lawyer using them is the accountability layer that makes the tool usable, not a bottleneck to be engineered out. That tension will not resolve because a press release chose a particular verb, and ACM is worth watching with that in mind.

Habeas is built for Australian legal practitioners who need research they can verify and own. Book a demo at habeas.ai.

The legal research in this article was conducted and every citation verified using Habeas, the Australian legal AI research platform.

Hero image: Mediamodifier on Unsplash

Other blog posts

see all

Experience the Future of Law