"Once information leaks, the damage is irreversible" — a new AI design built specifically for legal work

This article is an English translation of an article originally published on BUSINESS LAWYERS. Translation by BoostDraft.

Photo by BUSINESS LAWYERS

Contents

1. Beyond “AI for AI’s Sake”: Challenging the “AI Solves Everything” Myth
2. The Less Noise, the Better: Simple AI Review, Honed to the Essentials
3. Work Seamlessly in Word: Why BoostDraft Built It In-House
4. “We Don’t Train on Your Data” Isn’t Enough: A Design That Keeps Confidential Information Away from AI
5. Legal Talent Splits in Two in the AI Era: Judgment, the Creative Value That Remains with Humans

BoostDraft is a legal document editor that runs inside Microsoft Word. It has focused on eliminating tedious formatting work, such as adjusting indentation, catching inconsistent terminology, and checking clause numbers and cross-references. By deliberately choosing not to offer substantive review through generative AI, the company carved out its own position and earned strong support from large corporations and law firms.

In 2026, BoostDraft launched BoostDraft AI, which uses AI to perform a first-pass review of a document’s substance. Why move into AI review after all? Behind the decision was a growing concern that adopting and using AI was becoming an end in itself, along with the company’s own answer to how AI should be used in legal work: combine AI, rule-based technology, and human judgment, each where it fits best, while keeping confidential information from ever leaving the organization. That combination, the company believes, is what makes truly powerful legal tech.

Hiroshi Watanabe, Founder and Advisor of BoostDraft, spoke about the design philosophy behind BoostDraft AI and the value he believes legal professionals will continue to bring in the age of AI.

Beyond “AI for AI’s Sake”: Challenging the “AI Solves Everything” Myth

In February 2026, BoostDraft moved to a new management structure under Sean Yu as CEO, who has driven product growth at a number of global technology companies. Hiroshi Watanabe, who had led the company as co-founder and Chief Revenue Officer (CRO), stepped into the role of Founder and Advisor, a change that allows him to focus on product development, where his strengths lie and where his passion runs deepest.

BoostDraft AI was released under this new leadership. So why would BoostDraft, a company built on streamlining the formatting side of legal work, take the leap into AI review?

Watanabe describes what drove the decision as "frustration at how, in so many organizations, adopting and using AI has become an end in itself."

In the first half of 2026, reports that Big Tech companies such as Meta had begun factoring employees’ generative AI token consumption into performance reviews^[1] drew widespread attention.

These moves came amid a “just use AI, no matter what” mindset, but company after company ended up footing enormous bills. Uber, for example, used up its annual budget for AI coding tools in just four months.^[2] Cases also emerged where AI use wasn’t translating into results at all, such as employees running unnecessary tasks simply to rack up tokens. As a result, the focus is now shifting: rather than driving up usage, companies are narrowing AI to where it’s genuinely needed and aiming for more value per token.

“Technology is meant to be nothing more than a means to solve problems,” Watanabe points out. “But somewhere along the way, using AI, which should have been the means, turned into the goal itself. People were applying generative AI even in areas where it’s weak, or frankly, where it isn’t needed at all.”

Something similar is now beginning to happen in Japan’s legal sector. In the rush to “solve everything with generative AI,” some organizations are forcing it into areas where it’s weak, pouring enormous time and money into output that’s of little use in actual practice. It’s the same mistake made during the digital transformation boom a few years ago, all over again.

BoostDraft has long stood apart from AI review tools, but that never meant it was against generative AI. Quite the opposite: the company has consistently been optimistic about what generative AI can do. At the same time, it has always been keenly aware of one fact: AI is by no means a cure-all.

In legal documents, some work is best left to generative AI, while other work is better handled by rule-based processing that follows specific conditions and rules. Draw a clear line between the two, put each where it fits best, and you get the greatest efficiency. BoostDraft AI is built on exactly that concept, and in that sense, it stands as a direct challenge to today’s “AI solves everything” thinking.

Hiroshi Watanabe, Founder and Advisor, BoostDraft Inc. / Attorney-at-Law

Photo by BUSINESS LAWYERS

The Less Noise, the Better: Simple AI Review, Honed to the Essentials

In legal practice, where does the line fall between what work should be left to AI and what should not? Watanabe puts it this way: "AI is best kept out of work where mistakes aren’t an option and where the processing must never change what a document means."

Take, for example, asking generative AI to clean up inconsistent wording in a contract, such as conjunctions that are often used interchangeably in non-legal contexts when they shouldn’t be in legal contexts. Contracts can run long, and running one through generative AI burns through a huge number of tokens and leaves you waiting for the output. And on top of all that, there is a risk that hallucination changes the meaning of the text itself. Formatting work that leaves no room for variation should be done quickly and accurately, and there’s simply no need for generative AI there.

Generative AI shines, on the other hand, in areas where some range in the output is acceptable. Have ten top lawyers review the same contract, and each will flag slightly different points. Areas where more than one view can be valid are exactly where the strengths of generative AI come through most, according to Watanabe.

The division of labor, then, is clear: AI for reviewing substance, rule-based processing for formatting. So how does BoostDraft AI, which handles the substance, differentiate itself from existing AI review tools?

According to Watanabe, many existing AI review tools tend to compete on how detailed their comments are and how many features they offer. A broad feature set may look convenient at first glance, but in practice it can create additional work and defeat the purpose, with users saying, "The AI comes back with so many long comments that just reading through them is exhausting and eats up my time."

That’s why BoostDraft AI deliberately takes a “less is more” approach, paring its features down as far as possible. Only what users truly need remains, honed specifically for legal practice.

Of course, the BoostDraft team has taken great care in deciding which features to keep and how to calibrate the level of detail in its comments. BoostDraft has many lawyers and experienced legal professionals who have worked on the front lines. They have tested the product repeatedly against real-world workflows, debating it rigorously, with remarks like "this many flags actually makes it harder to use."

Watanabe offers a candid look inside the team: “Actually, quite a few of them are AI skeptics who believe that, at least for now, human legal professionals are far better than AI.” The product has been refined to the point where even these professionals, with their critical eye, admit that "this is actually useful in everyday work."

Another defining feature is a development philosophy fundamentally different from that of tools that tend to emphasize the sheer volume of comments and flags. When there are too many flags, users are left unsure what to check and how to decide. BoostDraft AI deliberately keeps its features and buttons to a minimum, simply highlighting the key points users should look at and leaving the final judgment to people. Generating as many review points as possible isn’t the goal. What matters is whether the tool helps users make decisions with confidence and complete their final work product.

Work Seamlessly in Word: Why BoostDraft Built It In-House

The decision to build in-house, rather than integrate another company’s AI review tool, also came down to usability in day-to-day practice. “To integrate deeply with BoostDraft’s core features and make it work as if it were a standard part of Word, building it ourselves was the only option,” Watanabe says.

Integrating a third-party tool would mean leaving Word, where the work starts, opening a browser, uploading the file, waiting several minutes for results, and then manually carrying them back into Word. With BoostDraft AI, users simply click a button in the same Word window they use every day. The AI suggests revisions, and if users want to apply them, one more click inserts them into the document as tracked changes. There’s no need to leave Word, and no need to change existing workflows.

BoostDraft AI also offers summarization and translation, but these work differently from typical generative AI tools. Rather than simply feeding a document into an LLM (large language model) and returning the output, they’re connected to the rule-based features BoostDraft has built over the years. If a summary mentions clause numbers or defined terms, users can check the relevant clause or definition in one click using the Reference feature. Translations, too, preserve contract-specific formatting and the correspondence between defined terms as much as possible. According to Watanabe, these capabilities are only possible because generative AI has been integrated with BoostDraft’s existing technology for preserving a document’s format and structure.

“We Don’t Train on Your Data” Isn’t Enough: A Design That Keeps Confidential Information Away from AI

Security is another major concern for legal departments adopting generative AI. Until now, BoostDraft has addressed this by processing everything locally, with no internet connection required.

In recent years, however, "shadow AI" — employees using generative AI and similar tools outside the company’s oversight — has become a growing problem, and many companies remain wary of using generative AI, or any tool with comparable capabilities, at all. While it’s now standard for providers not to use input data to train their models, that hasn’t been enough to dispel these deep-rooted concerns. This is especially true for legal departments: the contracts they handle every day are exactly the kind of information where, once it leaks, there’s no undoing it.

Even with models that don’t train on your data, the contract itself still has to be handed over to the AI. At security-conscious companies, that often means someone has to anonymize the data by hand before it can be used.

That’s why BoostDraft decided to go a step further than the industry-standard “we don’t train on your data” by introducing a mechanism that anonymizes the data sent to the AI itself. Users give instructions in a chat window built into Word, and with the anonymization feature turned on, confidential information such as company names, addresses, and dates is automatically anonymized before it reaches the AI. The AI’s response is then displayed with the masked details restored to their original names. To the user, it feels like any other exchange, but what the AI actually receives is anonymized data, so the masked details never reach the AI.

This approach reflects a perspective shaped by a team with many members who know firsthand the tension of contract work, where once information leaks, there’s no way to undo it.

Under the premise that "people, by default, don't want to change from the way they are used to," BoostDraft has worked to keep the barrier to adoption as low as possible. That means designing products that do not force users to change their workflow, and developing products that keep data in users’ own hands for security reasons. This philosophy, rooted in behavioral economics and behavioral science Watanabe studied during his MBA, carries through BoostDraft AI as well.

Legal Talent Splits in Two in the AI Era: Judgment, the Creative Value That Remains with Humans

As the legal landscape changes dramatically, with developments such as Anthropic’s release of “Claude for Legal,” what kind of added value will legal professionals need to bring going forward?

Watanabe answers without hesitation: "Thinking through unprecedented situations, and making the final judgment." The reason, he says, is that today’s generative AI "cannot carry out truly creative work."

Watanabe says he uses generative AI twenty hours a day, seven days a week, and that experience has shown him that "there is a surprising number of tasks AI cannot do." Ask generative AI to come up with several ideas, for example, and you’ll often find that every one of them feels like something you have seen before. Only human creativity can weigh those ideas and raise their value.

"Ideas are worthless; execution is everything." Until recently, this was a widely held view in the business world, especially in Silicon Valley. From now on, generative AI will handle the execution. It’s the ideas that hold the value.

Applied to legal practice, generative AI can handle execution tasks such as "fix this," and even suggest revisions like "it might be better to change it this way." But choosing among those options, deciding which risks to take and which to avoid for one’s own business, is a creative act only humans can perform. “That’s where the core value of legal professionals will remain,” Watanabe says.

Watanabe predicts that legal talent will divide in two. On one side are those who do little more than keep feeding inputs into generative AI. On the other are top professionals who make the final call and create real added value. He expects the gap between them to keep widening.

The key to becoming one of those top professionals, Watanabe believes, is “not placing blind trust in AI.” When you have AI produce something, you shouldn’t simply accept the output and call it done. Instead, push back with questions like "Why did you suggest this revision?", "What risks are you anticipating?", or "Explain this so someone without specialized knowledge can understand it" — and deepen your own understanding through dialogue with generative AI.

“People’s creativity and judgment are sharpened as different kinds of knowledge and experience come together. Use AI not as something that thinks for you, but as a partner that lets you encounter many perspectives and ideas in a short time. Doing that again and again is what builds your ability to think.”

Finally, Watanabe shared his vision for BoostDraft's future. "We have no intention of pushing heavy, monolithic all-in-one solutions on anyone," he says. Every company’s legal function is organized differently and works in its own way. On that premise, rather than forcing a "fit-to-standard" approach where companies must adapt their work to the system, BoostDraft aims to let each company choose only the capabilities it’s missing from its product lineup. That, Watanabe believes, is what putting the right tool in the right place truly means.

“AI is not a cure-all. Rather than thinking you can just leave everything to AI, what matters is using AI, rule-based technology, and human judgment, each where it performs best. I hope BoostDraft AI can help legal teams put the right tool in the right place.”

1. NewsPicks, "Meta has formally introduced AI use into performance evaluation. This trend will spread to every company" (in Japanese): https://newspicks.com/news/16355920/body/
2. CNET Japan, "The 'using too much AI' problem Amazon also faces: what lay behind the wasteful token consumption that spread internally" (in Japanese): https://japan.cnet.com/article/35248268/

Sourse

BUSINESS LAWYERS（なぜBoostDraftはAIレビューに踏み込んだのか？「情報が漏れれば取り返しがつかない」法務領域ならではの新しいAI設計／https://www.businesslawyers.jp/articles/1570）