Part 2 — ‘Legal engineering’ Didn’t Begin with Generative AI

The recent enthusiasm for the term ‘legal engineering’ invites a historical question. When attention concentrates on what generative artificial intelligence enables lawyers to build, it can become easy to mistake a change in technological capability for the beginning of the underlying activity. Lawyers can use natural-language instructions to develop applications, automate workflows and demonstrate working prototypes and applications. These possibilities give renewed practical significance to the relationship between legal expertise and technological construction. Understanding what has changed requires examining what came before.

But the underlying intellectual problem is much older. For more than half a century, researchers have investigated whether legal rules can be represented formally, whether legal reasoning can be assisted by computers, whether documents can be generated through structured systems and whether legal knowledge can be translated into forms that machines can process. Their approaches differed considerably. Some worked with symbolic logic, others with information retrieval, computational models, expert systems or document automation. They did not necessarily understand themselves as members of a common field, much less one called ‘legal engineering’. Nevertheless, their work forms an important part of the historical context in which the contemporary field should be examined.

The history of computational approaches to law is not the same thing as the history of the possible occupation associated with the title ‘legal engineer’. Confusing the two would produce an artificially continuous story, in which today's professional identity is projected backwards onto researchers who operated under different assumptions, institutions and disciplinary traditions.

The distinction matters because the present debate involves two different questions. One concerns the historical development of technical approaches to legal work. The other concerns when, how and why the term ‘legal engineering’ emerged as a description of professional activity. The first history is extensive. The second is less settled. Understanding their relationship helps explain both the possibilities and the difficulties confronting the field today.

I. The 1950s: Legal rules as formal structures

A useful starting point is 1957, when Layman E Allen published ‘Symbolic Logic: A Razor-Edged Tool for Drafting and Interpreting Legal Documents’ in the Yale Law Journal.¹ Allen's subject was not a contemporary occupation under the title ‘legal engineer’. His concern was the use of symbolic logic to improve the drafting and interpretation of legal documents. The significance of this work lies in the proposition that legal language could be examined through formal structures rather than solely through ordinary linguistic interpretation.

Legal provisions frequently contain combinations of conditions, exceptions, alternatives and consequences. Their meaning depends not only on individual words but also on the logical relationships between propositions. Symbolic logic offered a means of making some of those relationships explicit. This was important because legal drafting often relies on natural language to express relationships that can be difficult to identify or evaluate consistently. A formal representation may expose ambiguities, contradictions or unintended consequences that remain concealed in prose.

Allen’s work placed the logical structure of legal language at the centre of the enquiry. Making a provision’s conditions and alternatives explicit could help a drafter or reader examine how its parts related to one another. The connection with later computational systems lies in that problem of representation: before a rule can be processed, someone must decide how its meaning and structure will be expressed. Allen’s contribution provides a historical and conceptual starting point for that problem.

Legal information retrieval addressed a related but distinct problem. As legal materials accumulated, researchers began investigating how computers might assist in locating relevant statutes, cases and other legal sources. These efforts raised problems involving classification, indexing, language and the relationship between a legal question and the documents capable of answering it. Information retrieval and formal logic were not identical undertakings. One concentrated on locating information; the other on representing relationships within legal propositions. But both challenged the assumption that legal work was inherently beyond systematic computational treatment. The foundations of what would later become a much broader field of legal technology were already being explored.

This is one entry point into the history, not its geographical centre. Work on legal information and computation also developed through European research communities whose institutional trajectories were not simply extensions of American legal AI. The historical question is how these different programmes emerged, interacted and acquired their own vocabularies, rather than which country first anticipated the contemporary ‘legal engineer’.

II. The 1960s: From formal logic to computational legal reasoning

During the 1960s, research increasingly considered the relationship between legal reasoning and computational methods. Allen continued developing his work on logic and legal language, while other researchers examined whether computers could assist with legal analysis and the organisation of legal information. This period should not be treated as the emergence of a single, unified technical discipline. Researchers approached the subject from different intellectual backgrounds, including law, logic, computer science and information science. Their objectives ranged from improving the precision of legal expression to exploring whether aspects of legal decision-making could be represented computationally.

The underlying questions were demanding. Could a legal rule be separated from the language in which it was expressed? Could the conditions governing its application be represented formally? How should exceptions, qualifications and conflicting rules be handled? What happened when legal concepts depended on interpretation rather than fixed factual conditions?

These questions exposed a tension that remains relevant today. Legal rules may contain structures that lend themselves to formal representation, but their application often depends on context, interpretation and judgment. A system capable of processing a precisely defined rule is not necessarily capable of resolving the legal questions involved in deciding what the rule means, whether it applies or how it interacts with other sources of authority. This does not diminish the importance of computational approaches. It identifies the limits within which particular approaches operate.

It also helps explain why the historical development of legal technology cannot be reduced to a steady progression towards the automation of lawyers. Some projects sought to support legal reasoning. Others attempted to formalise particular legal tasks. Still others concentrated on the organisation and retrieval of information. The distinction between assisting professional work and reproducing professional judgment was present long before contemporary discussions of generative AI.

In Sweden, Stockholm University traces the activities of its Swedish Law and Informatics Research Institute to 1968. The institutional tradition of rättsinformatik illustrates a distinct point of departure: law and information technology could be studied through legal education, research methods and the organisation of information systems, rather than exclusively through attempts to simulate legal reasoning. This institutional account establishes a research lineage, not the historical existence of an occupation under the title ‘legal engineer’.²

III. The 1970s: Artificial intelligence and the possibility of computational legal systems

By the 1970s, the relationship between law and artificial intelligence had become a more explicit subject of research. In 1970, Bruce Buchanan and Thomas Headrick published ‘Some Speculation About Artificial Intelligence and Legal Reasoning’ in the Stanford Law Review.³ The article is an important historical marker because it addressed the possible relationship between developments in artificial intelligence and legal reasoning at a time when the technical and institutional circumstances were very different from those of today. The central issue was not whether a lawyer could quickly produce an application. It was whether computational methods might be capable of representing or assisting forms of reasoning involved in legal work.

A. Different questions and technical programmes

One way into the contemporary debate is through the accessibility of tools: what an individual can build, how quickly an application can be produced and how technical construction might change the economics of legal services. Earlier AI-and-law projects also make it possible to begin with the intellectual task itself. What constitutes a legal rule? How are facts represented? How are legal conclusions derived? Which elements of legal reasoning can be formalised, and which depend on interpretation or discretion? The technical possibilities and the underlying legal questions need to be examined together. The earlier questions remain relevant whenever a contemporary system claims to apply legal rules, classify legal materials or generate conclusions from structured inputs.

During the decade, researchers developed systems intended to address more particular legal tasks. JUDITH, TAXMAN and LEGOL illustrate different approaches. Sprowl’s account describes JUDITH as accepting normalised statutes, asking questions and drawing legal conclusions. TAXMAN applied computational representations of concepts to questions about the taxation of corporate reorganisations. LEGOL explored a formal language for expressing legislation and the information systems it could define. These projects pursued different technical objectives. Together, they show researchers investigating how legal knowledge could be represented and used in executable systems, alongside work on information retrieval and drafting.⁴

A parallel institutional history was taking shape in Belgium. The University of Namur dates the Centre de Recherches Informatique et Droit (CRID) to 1979. Its retrospective account describes an interfaculty structure bringing together law, computing and economics, alongside external professional participants. The initial research topics included judicial decision support, privacy and computing contracts. The centre provided an institutional setting in which different disciplines could investigate the relationships between law and computing over time.⁵

B. From representing legal knowledge to producing legal documents

An especially instructive development appeared towards the end of the decade. James A Sprowl’s 1979 article, ‘Automating the Legal Reasoning Process: A Computer That Uses Regulations and Statutes to Draft Legal Documents’, described the American Bar Foundation’s work on a processor for assembling legal documents. The abstract explains that attorneys or paraprofessionals could express documents and instructions in a language based on a subset of English, without conventional programming. The processor would ask questions, perform computations and produce documents adapted to the client’s circumstances. The relationship between legal knowledge, a usable authoring language and practical outputs was already an explicit design concern.⁶

Generating legal documents through computational processes involves decisions about how legal rules should be represented, how relevant facts should be collected and how outputs should be determined. A document is not necessarily a simple collection of interchangeable clauses. Its content may depend on legal conditions, factual circumstances and relationships between provisions. Translating those dependencies into a functioning system requires decisions about both the legal subject matter and its technical representation.

The historical significance extends beyond the existence of document automation. Sprowl’s account joins several forms of work: analysing legal dependencies, designing a language, collecting facts and controlling document production. It also shows an attempt to let people with legal knowledge construct systems without writing conventional programs. The processor’s capabilities were specific to its design and purposes. Its relevance to contemporary debate lies in the questions it addressed about who could build, how legal expertise could be represented and how a system could turn that representation into a useful output.

IV. The 1980s: Expert systems and the organisation of legal expertise

A. Representing expertise

The development of expert systems brought a further dimension to the relationship between law and computation. An expert system seeks to represent knowledge within a defined domain so that a computer can assist in addressing problems associated with that domain. In legal settings, this raised questions about the representation of rules, the identification of relevant facts and the production of conclusions or recommendations within particular areas of law.

Donald Waterman and Mark Peterson’s 1981 report, Models of Legal Decisionmaking: Research Design and Methods, forms part of this history. Their work on legal decision-making and expert systems explored how legal expertise might be organised in computational form.⁷

The significance of such work is not that computers had somehow acquired the full professional capacities of lawyers. Rather, expert-system research required investigators to confront the structure of expertise itself. Knowledge that may be applied intuitively or through professional experience must be identified, articulated and represented in a manner that a computational system can use. This process can expose the complexity of apparently routine decisions. It can also reveal how much professional knowledge depends on tacit assumptions, institutional context and judgments that are difficult to express as fixed rules.

The problem of translating expertise into systems remains central to many contemporary applications. A lawyer building an automated workflow, a software engineer developing a legal reasoning tool and a multidisciplinary team implementing a legal AI product may all confront versions of the same underlying difficulty: how to represent legal knowledge without losing the qualifications and context that give it meaning. The technologies differ, as do their capabilities and limitations. The underlying challenge has a recognisable history.

In Germany, Leibniz University Hannover dates the research unit preceding its Institut für Rechtsinformatik to 1978 and the formal establishment of the institute to 1983. Its account describes collaboration among lawyers, computer scientists and economists, and frames Rechtsinformatik as applied informatics concerned with the theoretical and practical consequences of information technology for law. The institution provides another setting in which legal knowledge, technical methods and professional practice could be examined together.⁸

B. Combining and distributing expertise

The expert-systems period also highlights a question that those working in the contemporary field cannot avoid. Who performs the work of translating legal expertise into computational form? The answer was not necessarily a lawyer acting alone. Such systems could involve collaboration between legal specialists, computer scientists, knowledge engineers and others responsible for representing and implementing domain knowledge. The allocation of responsibility varied according to the project.

A lawyer who personally builds technology combines legal and technical expertise within one person. The historical examples also involve arrangements in which legal knowledge and technical implementation are distributed among different participants. A system may depend on the knowledge of a legal specialist, the design of a computational language and the work of those implementing the processor. Examining those contributions makes it possible to consider individual and collaborative arrangements without assuming that either supplies the universal model for the field.

The historical evidence instead points towards a recurring organisational question: whether the relevant knowledge and capabilities are combined within an individual, distributed across a team or embedded within an institution. That question has not disappeared with the arrival of generative AI.

These continental European histories also complicate a chronology organised solely around successive generations of software. Stockholm, Namur and Hannover developed institutional homes in which questions about law, computing, information systems and professional practice could be pursued over time. Their agendas were not identical, and their existence does not establish a unified European tradition. They do, however, show that the relevant history includes institutions, disciplinary collaborations and educational practices, not only individual systems or landmark publications.

In Sweden, Stockholm University reports that rättsinformatik became a compulsory component of legal education in the early 1980s. That development is relevant because it shows how a law-and-computing relationship could acquire a curricular place without creating an occupation called ‘legal engineer’. The distinction between institutionalising a subject and recognising a professional title is central to the history traced here.⁹

V. From research programmes to organisational systems

The movement from research programmes into organisational systems did not occur at one identifiable moment. Document automation, knowledge management and information systems created settings in which legal knowledge and technical work could be combined without a shared occupational title.

The important historical distinction is between resemblance and identity. A technical specialist configuring a document system or a lawyer organising legal knowledge may have performed activities resembling contemporary roles. That resemblance does not establish that either belonged to the same occupation.

The research community also developed durable international connections. The first International Conference on Artificial Intelligence and Law was held in Boston in 1987. Later conferences included Oxford in 1991, Amsterdam in 1993, Edinburgh in 2003 and Bologna in 2005. These gatherings provided recurring settings for research across law and computing. They help connect the earlier systems to the subsequent decades through an identifiable network of institutions and researchers, rather than an assumed succession of technologies.¹⁰

Organisational applications developed along a related path. In his 2010 retrospective, Richard Susskind describes work with Phillip Capper on a legal expert system addressing limitation periods, documented in their 1988 book, followed by his engagement with litigation support, knowledge databases and work-product retrieval. He also describes a later shift towards web-based legal guidance and the wider distribution of legal knowledge. This account supplies examples of changing applications and delivery arrangements between the expert-systems period and the later occupational accounts. It brings attention to the work of organising expertise for use by others, alongside the construction of individual systems.¹¹

Occupational history therefore requires contemporaneous job descriptions, work accounts, institutional records and documented uses of titles, alongside the history of particular systems. The dated accounts examined next provide anchors for that enquiry rather than a claim to complete the intervening chronology.

VI. The 2010s and the emergence of occupational language

A. Dated uses of occupational language

By the 2010s, the title ‘legal engineer’ had become visible in contemporary legal technology discourse. This is an important development because it marks a shift from examining activities that resemble present-day practice to examining explicit uses of the occupational language itself.

Two dated publications provide more precise anchors, while serving different evidential purposes. Susskind’s 2010 article, an edited extract from The End of Lawyers? (2008), envisaged legal knowledge engineers who would organise and restructure legal knowledge for incorporation into systems. It records a proposed role. In August 2016, Drew Winlaw described his own work as a legal engineer at Wavelength, involving data, process improvement, system design and automation. He also referred readers to Stuart Barr’s HighQ article, “The rise of the legal engineer”. Winlaw’s account records occupational language being used to describe actual work. Neither publication establishes the first use of the terminology.¹²

These sources are not all equivalent. Some concern the use of the title ‘legal engineer’. Others describe organisations, participants or roles associated with technological innovation in legal services. Still others offer retrospective accounts of earlier ideas. A retrospective statement about when a term was first conceived is different from a contemporaneous document demonstrating that the term was being used in a particular professional setting. The distinction is especially important when considering claims about who originated the field.

An individual may have used the phrase early, promoted a particular interpretation or helped establish an occupational role. Those are potentially significant contributions. They are not necessarily the same as creating the underlying practices, founding a profession or establishing a definition accepted across the field.

B. From description to professional identity

The emergence of a job title changes the nature of the historical enquiry. When an employer recruits someone under the title ‘legal engineer’, the term begins to perform an occupational function. It identifies a position, signals expectations and potentially creates a pathway through which others encounter the designation. When a business uses the title ‘legal engineer’ for employees, the title may also become part of its commercial identity. These developments are different from the publication of research into computational legal reasoning. They involve the organisation of professional labour.

But occupational recognition within individual organisations is not equivalent to recognition across a profession. Different employers may use the same title for different responsibilities. They may require different qualifications, place different emphasis on legal and technical experience, and provide different forms of career progression. The occupational identity can therefore become more visible without becoming standardised. That is precisely the problem identified in the first paper of this series. The historical record helps explain why. The field did not emerge from a single academic discipline, professional body or technical methodology. Its contemporary activities developed against a background of multiple research traditions and legal technology practices. The later adoption of a shared term does not erase those differences.

C. Organising workflows before generative AI

The organisational history also includes legal operations. In a 2019 ACC account, Catherine Moynihan and Rachel Zahorsky describe earlier work on automation, legal intake, process analysis and the allocation of legal resources. Their retrospective examples include the work of VMware’s legal operations leadership and SolarCity’s redesign of legal service intake. This is evidence of organisational practices before the present generative-AI movement, rather than proof of one continuous occupation.¹³

The continuity concerns problems and responsibilities: organising work, representing information and deciding where human judgement is required. New tools can change how solutions are produced and who can implement them. Keeping those changes distinct from the earlier organisational needs helps avoid treating a new means of building as the invention of the entire activity.

VII. The 2020s: Generative AI changes the conditions of participation

A. New means of building

Generative AI changes the practical circumstances in which legal professionals encounter software development. Natural-language interfaces and AI-assisted programming can help users explore interfaces, workflows, document-processing tools and other applications without first mastering every part of a conventional development environment. A lawyer can describe intended behaviour, inspect an implementation and use the tool to assist further development. These possibilities create additional routes into technical experimentation, alongside collaboration with software specialists and the use of existing development platforms.

These routes sit within a longer history of attempts to make system construction accessible to people with domain knowledge. Sprowl’s document-assembly language is one earlier example; other approaches have involved configuring rules, forms and workflows within existing tools. The contemporary question is therefore more specific than whether lawyers can build without conventional programming. It concerns the flexibility of the tools, the range of tasks they can assist, the resources required to experiment and the expertise needed to turn a working prototype or application into a dependable system. Those questions can be investigated without assuming that a newly visible professional identity represents everyone doing related work.

But accessibility is not the same as historical novelty. The desire to translate legal rules into executable structures predates generative AI. So does the attempt to automate legal documents, model legal reasoning and construct systems incorporating legal expertise. What has changed is the technological environment in which some of those ambitions can be pursued.

There are also important differences between earlier systems and contemporary generative AI. Traditional rule-based systems generally depended on explicit representations of rules and knowledge within defined structures. Generative AI systems operate through different computational methods and can produce outputs without the same form of expressly encoded legal reasoning. These differences affect how systems are developed, evaluated and understood. It would therefore be equally mistaken to describe contemporary AI applications as merely updated versions of 1970s expert systems. Historical continuity does not imply technical equivalence. The point is that the ambitions, questions and professional relationships surrounding legal technology have deeper roots than the current generation of tools.

B. Building and professional identity

The growth of AI-assisted development also brings the distinction between activity and professional identity into sharper focus. A lawyer who builds an application may acquire valuable technical experience. They may produce a system that addresses a genuine problem. They may eventually develop substantial expertise in software construction and implementation. None of these possibilities should be dismissed.

But the ability to build an application does not, by itself, answer whether the individual belongs to a particular profession. That question depends on how the relevant professional category is understood, what evidence supports its boundaries and how its participants are recognised. The historical record cannot supply those answers automatically. It can, however, prevent the discussion from beginning with an artificial assumption: that the field emerged because generative AI made it possible for lawyers to build software. The relationship between law and computational systems is far older.

VIII. What the historical record establishes — and what it does not

The historical record supports a substantial history of computational approaches to law and a separate, more recent history of explicit occupational language. Neither history establishes a single professional lineage. The dated accounts examined here provide points from which further occupational research can proceed. They do not establish the first use of the title ‘legal engineer’, a single founder or the prevalence of particular responsibilities at any earlier date. The contribution of history is to identify the problems, institutions and collaborations that contemporary descriptions may otherwise leave outside the frame.

IX. The danger of starting history too late

This history matters to the contemporary debate because the starting point of a historical narrative can influence the understanding of a contemporary field. If the field is presented principally as a consequence of generative AI, attention naturally concentrates on the people and activities made newly visible by those tools. Lawyers who build applications become central to the story. The ability to code, configure or direct AI-assisted systems may appear to provide the defining characteristic of a proposed profession.

That is a legitimate subject of enquiry. But it is not the only possible account. A history beginning with formal legal representation, computational reasoning, expert systems and the development of legal technology practice directs attention towards a wider set of questions. How is legal knowledge represented? What technical work is required to construct dependable systems? How do legal and technical specialists collaborate? How are such systems evaluated, implemented and maintained? How have employers and participants described the work over time?

These questions do not necessarily produce a broader definition of the field. They demonstrate why any proposed definition should be tested against a broader body of evidence. The risk of starting the history too late is not simply that earlier contributors are overlooked. It is that the contemporary field may be understood through an unnecessarily narrow account of how its activities developed. Historical research cannot determine professional boundaries by itself. It can, however, expose assumptions that might otherwise pass unnoticed. That is the contribution of historical research to the present debate.

Conclusion An older problem, a newer identity

The problems associated with the term ‘legal engineering’ did not begin with generative AI. The historical record shows that lawyers, computer scientists and other researchers have spent decades examining how legal knowledge can be formalised, represented, processed and incorporated into functioning systems. Their work cannot be collapsed into a single discipline or retrospectively assigned a common occupational identity. Yet it provides essential context for understanding the technological and intellectual questions confronting contemporary participants.

The more recent emergence of the title ‘legal engineer’ raises a different set of issues concerning occupational identity, employer expectations and institutional recognition. Those questions remain open. Generative AI has changed the accessibility and visibility of technical construction. It has created new opportunities for participants and renewed interest in the relationship between legal knowledge and software development. But it has not erased the history that preceded it. Nor has it resolved which activities and competencies justify recognition under the title ‘legal engineer’.

The next paper will examine how the field is made through the relationships among participants, organisations, technologies, institutions and the accounts through which they describe their work. The occupational evidence will be examined in greater detail in Part 4. Before deciding what the field should become, we need to understand both its historical antecedents and the work now described as legal engineering, including related activities performed under other occupational titles.

Publication note

This series forms part of the Institute for Legal Engineering™ research programme. The Institute intends to develop professional standards, independent certification and training. The research examines the scope and justification for those activities; it does not assume that a single occupation, professional definition or certification model has already been established.

References

Layman E Allen, “Symbolic Logic: A Razor-Edged Tool for Drafting and Interpreting Legal Documents” (1957) 66 Yale Law Journal 833–879, especially 833 and the section “Application to Concrete Legal Problems”. Source

Stockholm University, Department of Law, “Centres and other units”, Swedish Law and Informatics Research Institute section (institutional retrospective, accessed 15 October 2026). Source

Bruce G Buchanan and Thomas E Headrick, “Some Speculation About Artificial Intelligence and Legal Reasoning” (1970) 23 Stanford Law Review 40–62, especially 40–41. Source

Walter G Popp and Bernhard Schlink, “JUDITH, a Computer Program to Advise Lawyers in Reasoning a Case” (1975) 15 Jurimetrics Journal 303–314, as described by James A Sprowl, “Automating the Legal Reasoning Process” (1979), note 20. Source

L Thorne McCarty, “Reflections on TAXMAN: An Experiment in Artificial Intelligence and Legal Reasoning” (1977) 90 Harvard Law Review 837–893, abstract in the author-uploaded publication record; see also Sprowl (1979), note 20. Source

R K Stamper, “The LEGOL 1 prototype system and language” (1977) 20 The Computer Journal 102–108, abstract. Source

University of Namur, “1978–1992: essor et développement de la recherche”, CRID section (institutional retrospective, accessed 15 October 2026). Source

James A Sprowl, “Automating the Legal Reasoning Process: A Computer That Uses Regulations and Statutes to Draft Legal Documents” (1979) 4 American Bar Foundation Research Journal 1–81, abstract and note 20. Source

Donald A Waterman and Mark A Peterson, Models of Legal Decisionmaking: Research Design and Methods (RAND, R-2717-ICJ, 1981), Summary, vii–viii. Source

Leibniz University Hannover, Institut für Rechtsinformatik, “IRI – damals und heute”, especially “Entstehung” (institutional retrospective, accessed 15 October 2026). Source

Richard Susskind, “Legal informatics – a personal appraisal of context and progress” (2010) 1(1) European Journal of Law and Technology, sections 3–4 and discussion of legal hybrids and legal knowledge engineers; edited extract from The End of Lawyers? (2008). Source

Drew Winlaw, “Are you a Legal Engineer?” (30 August 2016), first-person account of work at Wavelength, occupational terminology and relevant capabilities. Source

Catherine J Moynihan and Rachel M Zahorsky, “Legal Ops Can Transform the Practice of Law, but Only if Everyone Embraces It” (1 March 2019), ACC Docket 37(2), 44–51, “Dedication leads to transformation” and “Knowledge-sharing is natural”. Source

International Association for Artificial Intelligence and Law, “History”, ICAIL 1987, 1991, 1993, 2003 and 2005 entries (accessed 25 October 2026). Source