How Should Legal Workflows Change in the Age of Generative AI?
Hello, I'm Noriaki Asato, Representative Attorney at LegalAgent.
Have you introduced generative AI into your legal work, only to feel that the overall burden has not become as light as you had hoped? In such situations, it helps to look not only at the individual tasks you asked AI to perform but also at the internal work that arises before and after them.
Summarizing contracts, checking the meaning of provisions, and preparing proposed revisions and draft comments to the counterparty can now be handled with a certain degree of accuracy even by general-purpose generative AI. Even so, while the time needed to write a question and get an answer from AI has been shortened, the work of confirming the background of a request, researching past negotiation history, and tidying up tracked changes in Word remains.
Legal practice is not limited to writing text. It is work that involves opening a Word file, scrutinizing the counterparty's tracked changes, grasping the intent behind their comments, referring to the history stored in the matter folder, and presenting proposed revisions that reflect the business department's wishes. Unless these preceding and following steps are brought into view, generative AI remains a "convenient search box" and does not reach the point of fundamentally changing day-to-day workflows.
Working in the Browser and the Information That Does Not Get Through
When using generative AI, people usually paste contract provisions or the details of a consultation into a browser window and give instructions. This method can be useful at the stage of checking the general meaning of an individual provision, listing candidate legal risks, or getting suggested rephrasing. The accuracy of the explanations obtained should be verified separately.
In actual contract review, however, the Word file contains detailed tracked changes and comments from the counterparty. In addition, the business department provides negotiation background such as "this condition is non-negotiable under our business plan," "given our relationship with them, it is hard to push strongly," or "we have another large transaction coming up after this one." The moment you paste only part of the contract into a browser, the negotiation history and business context outside that text are not passed to the AI. As a result, even if the answer output by AI looks reasonable at first glance, it cannot be used as is in practice.
For contracts that have gone back and forth with the counterparty several times, judgment is required as to which comments to respond to, which tracked changes to keep, and which issues should be left to internal decision-making. As long as humans have to reconnect this surrounding context every time, introducing AI will not dramatically reduce the burden on in-house legal. In my view, the purpose of a legal workflow incorporating AI is not merely to increase the number of answers in the browser, but for AI and attorneys to engage with the context of the place where the matter is actually progressing.
Dividing Roles on the Basis of Word Files and Matter Materials
In workplaces where contracts are exchanged in Word, edits and comments on the file make up part of daily work. You revise provisions while keeping tracked changes, convey the purpose of revisions using the comment function, and respond appropriately to the counterparty's comments. At the same time, internally, you need to communicate concisely which issues are important and, where necessary, shape the draft so as to minimize unnecessary redlines.
What is required here is not limited to presenting legally accurate provisions. Choosing expressions that do not give the counterparty an overly adversarial impression, adding explanations that internal decision-makers can grasp intuitively, and appropriately managing the volume of tracked changes are also important practical elements.
Furthermore, matter folders accumulate past drafts, minutes of business meetings, and the history of what was debated in similar past contracts. Legal staff refer to these even without specific instructions to judge "how far we can concede as a company" and "what the business department is concerned about." Because such decision criteria tend to become tacit knowledge within in-house legal, if generative AI is to be built into practice, it is important to sort out how specific tasks will be divided within the Word files and matter materials.
Generative AI can produce clause proposals, revisions, and outlines of minutes as well-formed text. In legal practice, however, a neatly formatted draft cannot necessarily be applied as is. Even a single cap on damages cannot be decided uniformly from the form of the contract alone. The appropriate level varies with the size of the transaction, negotiating leverage relative to the counterparty, past incidents, and other factors. Even for a non-disclosure agreement, how the same provision is perceived differs greatly depending on the nature of the information received, plans for joint development, and whether an investment is being considered.
Therefore, a realistic division of roles is to use AI for preparatory work and have attorneys and legal staff add practical knowledge and perform the final check. AI handles extracting issues, cross-checking against past materials, and drafting comments, while attorneys verify those results and finish them into a usable form based on the company's decision criteria and negotiating priorities. With humans responsible for that check, AI is put to use in understanding and examining the matter.
Using Outside Attorneys Who Work Inside the Workflow
When using AI in legal work, think about the tools together with the work procedures. In my view, the significance lies in setting up a system in which outside attorneys enter day-to-day workflows in a position close to in-house legal and provide continuous support all the way through interactions with business departments and internal decision-making.
In traditional legal consultations, the common flow was for in-house legal to organize the issues, summarize the background, and clarify the questions before engaging an outside law firm. Even if this format yields high-quality answers from outside attorneys, the preliminary research before the consultation and the work of adapting the returned answer for internal use remain in-house.
With legal outsourcing that rethinks the workflow, there is more room for outside attorneys to take on these preceding and following steps as well. They directly review consultation memos and chat histories from the business department and past similar contracts, identify the issues, pose necessary follow-up questions to the business department, and present options aligned with internal decision criteria. If collaboration with outside counsel can extend to these steps, the burden of work previously kept in-house can be reduced. In this model, outside attorneys function not merely as people who answer legal questions but as collaborators who move internal decision-making forward.
In addition, at growing companies and startups, the volume of legal work fluctuates. There are periods when fundraising, partnership negotiations, and various regulatory responses are concentrated, and periods dominated by routine contract review. Trying to respond to these fluctuations solely by hiring internal staff creates a burden for recruitment and training, and fixed costs may become a heavy load during quiet periods. A mechanism that allows you to use legal capacity of the necessary depth at the necessary time as a variable cost is effective in responding to the ups and downs of workload.
In doing so, you want to avoid having to explain the background from scratch every time you consult. By sharing past matter folders, accumulated decision examples, contract templates, and the history of dialogue with business departments, and maintaining an ongoing relationship, even outside attorneys can proceed with work at a speed and accuracy close to those of in-house legal.
Designing Inputs and Output Formats
When rethinking business workflows, you need to decide in advance which information will be consolidated with the AI and attorneys, and in what form the deliverables will be returned internally.
As input information for contract review, gather in a single matter folder not only the contract itself but also past versions, comments from the counterparty, the business department's requests, and the history of similar past matters. With this information in place, it becomes easier for AI to present differences between provisions, candidate issues to be considered, and candidate items to confirm with the business department. Consolidating information requires some initial organization, but if you standardize storage locations and shared items in advance, subsequent matters can proceed smoothly using the same procedure.
As for output format, rather than returning a single long report, it is practical to return the output clearly divided into three: tracked changes in Word, comments for the counterparty, and an internal review memo. In particular, care is required not to confuse the reasons for revisions disclosed to the counterparty with risk judgments kept internal. To the counterparty you might say, "To clarify the scope of liability, we propose a provision setting a cap based on the contract amount," while internally you might write, "Under the current provision, there is a concern that we would bear broad liability even for minor negligence, so we will treat a cap provision as a negotiating option, taking into account its balance with the size of the transaction." Organizing and presenting deliverables in this way makes it easier for the business department to understand the negotiations and decisions it will carry out next.
Deciding even where in the Word document to leave comments, and for whom, makes the generated drafts easier to use in business. In setting up the procedure from request to delivery, too, the perspective of reviewing the overall flow affects results more than the choice of tools. There is no need to apply it to all contract types at once; a sound approach is to gradually expand the scope of application starting with the typical contract types that generate the most consultations.
Gather the contract, an overview of the matter, and related materials in a folder; have AI organize the main issues, candidate revisions, and questions; and then have attorneys or legal staff set the policy. In line with that policy, AI reflects initial drafts of tracked changes and comments in Word, and finally a human checks and finishes the deliverable. Once this procedure is established, legal staff can begin their work by reviewing the drafts prepared by AI, and can more easily spend their time on judgments such as "which risks should be prioritized in this matter" and "what response policy should we present to the business department."
Consultations on a Legal Structure Grounded in Practice
At LegalAgent, rather than replacing legal work with AI alone, we aim to change the entire legal workflow by combining Word, tracked changes, and attorneys' practical knowledge. If you want to speed up contract reviews and legal consultations, use outside attorneys in a way close to in-house legal, or build an AI-enabled legal structure that holds up in practice, please feel free to consult us.
Services connected to this topic