From Law as Code to Code as Law
AI is reshaping who controls the legal function. But when engineers write the rules, binary logic replaces legal judgement and your contracts may become a source of risk rather than value.
There is a quiet power shift underway in legal practice, and most businesses are moving too fast to notice it. The rise of AI legal tools, contract drafters, policy generators, clause comparison engines, is accelerating legal work in ways that feel like unambiguous progress. Faster turnaround. Lower cost. Consistent output. What is not to welcome?
Quite a lot, it turns out.
Lawyers have long recognised that law is a form of code, a structured system of rules, exceptions, and interpretive principles that govern how people and organisations must behave. But what AI is doing is inverting that relationship. It is not helping lawyers apply the law more efficiently. It is increasingly substituting the lawyer's judgement with the software engineer's logic. The result is a world where code becomes the law, and where the limitations of that code become the limitations of your legal protection.
1. The Illusion of Efficiency
AI legal tools are built on large language models trained on vast corpora of legal text. They can identify standard clauses, flag deviations from market norms, and generate first drafts at speed. For volume, repetitive work, non-disclosure agreements, standard supplier terms, routine data processing addenda, the efficiency gains are real.
But efficiency is not the same as quality. And in law, the gap between the two is where liability lives.
The fundamental constraint of AI-generated legal output is this: it operates on pattern recognition, not legal reasoning. It produces what is statistically probable given comparable documents, not what is legally optimal for your specific situation. A contract clause that appears in 90% of comparable agreements may be completely wrong for the 10% of circumstances that describe your deal.
2. Binary Logic Cannot Hold Legal Ambiguity
Law is not binary. It is a discipline built on the careful interpretation of language in context, statutory rules of construction, common law principles, the factual matrix surrounding a transaction, and the reasonable expectations of the parties at the time of contracting. Courts do not simply read what a contract says. They ask what it means.
AI systems cannot perform that exercise. They are trained on outputs, the finished legal text, not on the reasoning process that produced it. They have no access to the negotiation history, the commercial objectives that shaped each clause, or the regulatory backdrop specific to your industry and jurisdiction. They cannot apply the contra proferentem rule, weigh competing canons of construction, or anticipate how an arbitral tribunal in a particular seat might approach a damages cap.
This is not a temporary limitation waiting to be engineered away. It reflects something structural: legal interpretation requires contextual judgement, and contextual judgement is not reducible to pattern matching.
3. Standardisation as Risk
One of the selling points of AI-generated legal documentation is consistency. Every output follows the same structure, the same market-standard positions, the same drafting conventions. For businesses seeking to scale their legal function, this appears attractive.
But standardisation in law carries its own category of risk. Legal instruments derive their value not from conformity to a template but from their precision in capturing the specific rights and obligations of specific parties in specific circumstances. A contract that is 95% market-standard and 5% wrong for your situation is not a contract that provides 95% of the protection you need. It may provide none at all on the points that matter most.
Consider a few scenarios where AI standardisation generates exposure rather than protection:
Acquisition Agreements - AI may produce a standard MAC clause, but whether it captures the specific risk profile of this target business, its regulatory position, its customer concentration, its supply chain dependencies, requires bespoke drafting informed by due diligence findings.
SaaS & Licensing - Intellectual property ownership, escrow triggers, and service level remedies in AI-generated technology contracts may not reflect the actual commercial dynamics of the arrangement or the leverage each party holds.
Data Processing Agreements - Market-standard DPA may satisfy the surface requirements of UK GDPR Article 28. But the specific technical and organisational measures, sub-processor chains, and audit rights must be calibrated to the actual data flows and risk appetite, not a generic template.
4. Who Controls the Legal Function?
There is a deeper question here that sits beneath the technical debate about AI accuracy. When legal output is generated and deployed by engineers rather than lawyers, the locus of legal control shifts from those trained in legal reasoning to those trained in systems design. These are not the same skill set, and the difference matters.
A lawyer advising on a contract is not merely producing a document. They are exercising professional judgement, weighing risks, translating business objectives into enforceable legal rights, and accepting accountability for the advice given. That accountability is backed by professional regulation, insurance, and ethical obligations. An AI system has none of these. Neither does the engineer who deployed it.
This creates a governance gap that regulators are beginning to notice. Businesses using AI legal tools without adequate legal oversight are, in effect, making legal decisions without legal advice. The contracts they sign, the policies they publish, and the rights they purport to grant may not mean what they think they mean.
5. The Right Way to Think About Legal AI
None of this is an argument against legal AI. The tools are genuinely powerful for the right applications: document review at scale, precedent research, first-draft generation for standardised instruments, contract analytics across large portfolios. Used as an accelerant for legally-supervised work, AI can make lawyers significantly more effective.
The error is in treating AI output as a substitute for legal judgement rather than a prompt for it. The value of a contract lies not in its existence but in its enforceability, its capacity to protect you when the relationship it governs comes under strain. That enforceability depends on legal reasoning that AI cannot replicate.
Businesses that understand this will use AI to do more law faster, with qualified oversight. Businesses that do not will accumulate a portfolio of documents that look like contracts but function like liabilities.
6. Law as Business Value
The best legal work has always been a form of strategic advice, not just document production. A well-drafted contract allocates risk intelligently, anticipates dispute scenarios, and gives each party a clear understanding of their position, creating the conditions for a commercial relationship to succeed. That is law as business value.
Achieving it requires someone who can hold two things simultaneously: a precise understanding of legal principle, and a practical grasp of what the business actually needs. That combination, legal expertise applied to commercial reality, is what a skilled lawyer brings to a transaction.
AI can assist that process. It cannot replace it. The question for any business deploying legal AI tools is not "can we produce legal documents faster?" but "are the documents we are producing actually doing the legal work we need?" The answer to the first question is almost always yes. The answer to the second depends entirely on whether a lawyer is still in the room.
