Clients and law firms are all asking the same question: what AI tool should we be using? That’s the wrong question. Successful AI adoption is not about selecting the right tool: it requires a rewired organizational mindset and workflow design. Firms should revisit traditional workflows to develop multi-stage processes integrating different tools and appropriate human review. This will open up more complex projects to AI efficiencies. More importantly, it is essential to ensure accountability for AI outputs and mitigate regulatory concerns. The Organizational Mindset Law firms are increasingly realizing that integrated human and AI collaboration is the way forward. One residual barrier is cultural: the legal industry is built on evidence, proof and precedent, so it can have a binary viewpoint. This mindset says, “AI can hallucinate, so we shouldn’t use it.” Innovative litigators adopt a different mindset, embracing experimentation whilst understanding the need for rigorous evaluation. This mindset instead says, “AI can hallucinate, so we must have processes to verify it.” This mindset requires firms to reconsider their roles, systems, and processes. Roles shift from doing (drafting, researching), to managing (preparing context packages, setting clear prompts, and reviewing outputs). Attorneys must understand the mechanics of LLMs: how they work, what information to give them, and what their pitfalls are. This requires department-specific, hands-on training, and ongoing skills development, alongside real applications on live cases to reinforce learning. This also cannot come at a cost to legal skills and knowledge, which remain essential for effective human review. Junior attorneys now need a wider skillset than ever to bridge the gap between legal reasoning and prompt engineering. Building AI-Enabled Workflows Why workflows? To build workflows that truly integrate AI, firms should reconsider their processes from the ground up. A recent McKinsey report highlights that “the real value of AI is unlocked not by the technology itself, but by redesigning workflows to integrate AI capabilities at every stage of the process.” McKinsey; Agents for Growth: Turning AI Promise into Impact. Redesigned workflows have two main benefits: (1) Ensuring Ethical Compliance. A simple but vital reason: redesigning workflows will mitigate the risk of AI errors going unchecked. The ABA’s Opinion 512 requires lawyers using AI tools to maintain technological competence of the “evolving nature” of AI, including the “benefits and risks associated with relevant technology.” The Opinion reminds lawyers with managerial or supervisory responsibility that they must ensure that subordinate lawyers conform to the rules, including “clear policies” on the use of AI. ABA Comm. on Ethics and Pro. Resp., Formal Op. 512 (2024). (2) Optimizing Efficiency. The benefits are not only in compliance. AI—especially agentic AI—is immensely powerful when used properly. Ambitious companies which expect AI to transform their businesses will try to integrate AI as much as possible across their projects to reap the greatest rewards. Doing so requires them to view AI integration through workflow design, not through one-size-fits-all tools. Breaking it down. So, how should businesses redesign their workflows to best incorporate AI? Step one is to break down existing workflows into discrete steps, and for each step to consider the complexity and level of human interaction required. Workflow stages that involve strategy or judgment require more human interaction than document review and analysis, so should not be treated the same. Different tools. Likewise, firms should consider which AI tool is best for each workflow stage. One of the main concerns with AI is that platforms can hallucinate information. When you need deep research, the tool needs to access the entire internet to work, but this increases the chance of errors. If you want to analyze court filings, allowing the tool open-web access could do more harm than good. Here, a better tool might use Retrieval-Automated Generation (RAG). RAG uses a closed-universe of context, relying only on documents in your firm’s ecosystem and reducing the risk of hallucination. When rebuilding workflows for AI, think about the right tool or agent for the job. Human-in-the-Loop. The most important stage in workflow redesign is establishing how the AI output will feed into the overall work product. This should involve stringent review protocols and ethical safeguards for content accuracy and regulatory compliance. To prevent errors that could result in sanctions, legal workflows should ensure Human-in-the-Loop (HITL) architecture. HITL does not mean that lawyers will have to spend just as long reviewing as they would have spent drafting. Workflows can also include AI-led quality control, such as using a second AI tool to review the output of the first, or requiring the tool to assign confidence scores to its outputs to prioritize and triage human review. Before filing a brief or sending a memorandum to a client, human interaction to review AI output is essential. AI in US Litigation and Investigations Some US litigators have learned about AI risks the hard way. The industry was first warned back in 2023, after a plaintiff firm was hit with a $5,000 fine for using AI to generate fake citations in an S.D.N.Y brief. The message was clear: attorneys retain a gatekeeping obligation to ensure the accuracy of their filings, regardless of who (or what) drafted them. Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023). Two years later, the lesson is even more stark. In July 2025, a U.S. District Judge sanctioned three attorneys for fabricated citations in a discovery motion. The attorneys worked for a national law firm which had an AI use policy requiring independent verification of AI outputs. The court imposed a harsher sanction than previous reprimands, “account[ing] for the danger that fake citations pose” to the judicial system. The judge sanctioned not only the attorney who drafted the motion, but also the supervising attorney for failing to properly review the draft. Johnson v. Dunn, 792 F. Supp. 3d 1241 (N.D. Ala. 2025). This shows two things. First, don’t assume that the people you are working with fully understand the risks of AI. Having concrete workflows, including clear AI policies, is necessary, but will not be a shield if people do not apply them in practice. Second, it is no longer an excuse to be ignorant of AI risks. As the ABA’s Opinion says, technical competence of AI is “not a static understanding.” As AI becomes commonplace and agents develop, firms must ensure that all their attorneys know the capabilities and limitations of those tools. Key Takeaway Successful integration of AI should have mindset and workflow redesign at its core. As recent sanctions show, the risks of overreliance on AI tools are real. But these risks are products of flawed processes, not imperfect technology. By shifting the organizational mindset and redesigning processes from the ground up, firms can fully harness the power of AI while maintaining the ethical standards the bar requires. This is iterative: firms should not expect AI tools to immediately replace human judgement. But by treating workflow design as a strategic priority, and being guided by evidence of AI capability, you will get ahead without compromise. Tim Harkness, Rob McCallum and Eric Bruce are partners at Freshfields LLP. Cameron MacDonald is a trainee associate with the firm.