I delivered these remarks to a group of practicing attorneys in Ohio. You’ll see some remarks tailored to that state’s legal community. That said, the majority of the points made should have some value to lawyers across the U.S.
If you or any group you’re a part of would like to chat further, let me know! kevin.frazier@law.utexas.edu
Good afternoon. Big thanks to Jane Doe for the kind invitation. My name is Kevin Frazier, and I'm the AI Innovation and Law Fellow at Texas Law.
We stand at a precipice. Not just in technology, but in the very practice of law. Artificial intelligence isn't just knocking at the door; it's already remodeling the house. The question isn't if AI will fundamentally change our profession, but how profoundly, how quickly, and most importantly, how your firm will adapt not just to survive, but to thrive.
Today, I want to talk about "The AI Imperative." This isn't about chasing shiny new tech toys. I’m not selling anything!!
It's about recognizing a fundamental shift and strategically positioning your practice for a future that's arriving faster than many of us anticipated. We'll explore why the AI we see today is just the beginning, how the current friction against its use will inevitably ease, and crucially, how firms need to think beyond simply using AI tools to fundamentally revising systems, structures, and cultures around them.
Let's start with a crucial, perhaps counterintuitive, point: The AI you might be experimenting with today – ChatGPT, CoCounsel, Harvey, Claude – impressive as it might seem, is the worst AI you will ever use.
Think about that. We're seeing capabilities that were science fiction just a few years ago: drafting memos, summarizing depositions, identifying contractual risks, answering complex legal questions with citations.
We see models generating code, analyzing images, even assisting in predicting case outcomes based on historical data. And yet, the pace of improvement is relentless. The models are doubling in size and capability roughly every six months to a year. I was just at Meta’s Open Source AI Summit and I can tell you from the front lines that we haven't scratched the surface of what’s to come.
Today's large language models still make mistakes. They "hallucinate" – invent facts or cases. We’ve all heard the stories of lazy lawyers neglecting to check their cites and being scolded by their judge and colleagues!!
Their reasoning can sometimes be shallow or lack specific jurisdictional nuance without careful prompting.
Their knowledge may cut-off dates. I wouldn’t rely on some models for the latest news. Yes, there’s still very much a need for traditional journalism!
But AI’s trajectory is clear and steep. The underlying models are getting exponentially larger, trained on more diverse and specialized data (including legal-specific datasets), and refined with better techniques like reinforcement learning from human feedback specifically tailored for legal accuracy. Hallucinations are decreasing. Accuracy is increasing. Capabilities are expanding.
The AI of tomorrow will be significantly more accurate, more nuanced, capable of more complex multi-step reasoning, and more seamlessly integrated into specialized legal tasks and workflows. Waiting for a "perfect," error-free AI before engaging is like waiting for the perfectly safe car before learning to drive. Try telling your kid that they have to wait for an accident proof car before they get their permit. Not going to happen.
You'll be left behind. The imperative is to learn, adapt, and build processes now, understanding that the tools themselves will only get better, faster than we might imagine.
Now, I know what some of you might be thinking. "There's resistance. Courts are issuing standing orders... Clients are wary... Ethical concerns abound." You're right. There is friction. But I contend that today's legal ecosystem represents the peak friction AI will face.
Consider the advent of the automobile. When cars first appeared, they were technically superior to horses for personal transport in many ways – faster, more powerful. But the world wasn't ready. There were few paved roads, no gas stations, no traffic laws, no driver's licenses. People were scared of these noisy, dangerous machines.
Early adopters faced immense hurdles, ridicule, and regulatory roadblocks. Red flag laws in the UK required someone to walk six hundred feet ahead of the car waving a flag!
Did that stop the car? No. Because the value proposition was too compelling.
What happened? We built the infrastructure – roads, highways, gas stations. We created new rules and norms – traffic signals, speed limits, licensing, insurance. The ecosystem adapted to unlock the full potential of the technology.
We are in a similar moment with AI in law. The initial flurry of standing orders demanding disclosure, often born from uncertainty and high-profile examples of misuse (like citing fake cases), is already receding in many jurisdictions or becoming more nuanced. Why? Because courts and bar associations are realizing that blanket restrictions are impractical and that the focus needs to shift from whether AI is used to how it's used responsibly, competently, and transparently when necessary.
The "infrastructure" for AI in law is being built right now. Secure, enterprise-grade platforms with better data privacy controls. Integration with existing legal tech stacks (case management, e-discovery). Development of best practices and ethical guidelines by bar associations. New norms around verification, oversight, and client communication. Just as roads and rules unlocked the automobile, this evolving legal tech infrastructure and refined professional standards will smooth the path for AI integration. The friction you feel today is temporary; the underlying advantages in efficiency, capability, and even insight generation are permanent and growing.
Some members of our profession, however, have been slow to adjust to the inevitability of better AI, of ubiquitous AI.
I won't ask folks to raise their hands, but if we did an informal survey of this room, then my hunch is only about a third of you regularly use AI in their practice. A recent ABA poll found that just 30 percent of lawyers used AI in 2024. Now that's a big jump from 11% in 2023 -- in just one year, use of AI tripled, but it still suggests great hesitancy.
The massive surge in adoption as well as the rapid progress in AI suggests that 2025 will see ever more lawyers lean into AI. My goal today is to help you do just that.
So, if AI's advance is inevitable and the friction is temporary, how do we proactively prepare? It requires more than just buying a subscription to an AI tool. It demands rethinking core firm processes. Let’s start with hiring.
How many of you currently ask potential hires – from summer associates to lateral partners – about their experience with generative AI? If not, you should start tomorrow. This isn't about finding "AI wizards"; it's about assessing adaptability, critical thinking, technological competence, and frankly, their awareness of the profession's trajectory.
Consider adding questions like these to your interviews:
"Describe your familiarity with generative AI tools relevant to legal practice..."
"Can you give an example of how you've used an AI tool to assist with legal research, drafting, or case analysis? What specific task were you trying to accomplish? What prompt did you use? What were the results, and how did you verify them?"
"Tell me about a time an AI tool provided incorrect, incomplete, or misleading information. How did you identify the error, and what steps did you take to correct it? What did that experience teach you about using these tools?"
"Which AI tools do you prefer and why? What are their strengths and weaknesses in your view?"
The answers reveal more than just technical skill. They show whether a candidate is curious and ready to contribute to our collective effort to adjust to the Age of AI. Hiring lawyers who are AI-aware, critical, and adaptable is the first step in building an AI-ready firm.
Hiring adaptable people is crucial, but it's not enough if they enter a culture resistant to change or experimentation. True integration requires intentionally shaping your firm's structure and norms to encourage and normalize the use of AI. This means moving beyond top-down mandates and fostering bottom-up learning, experimentation, and crucial knowledge sharing.
How do you do this? Create regular, structured, and safe opportunities for your team to share their experiences with AI – both the wins and the warnings.
Practice Group Meetings: Dedicate 10-15 minutes in regular practice group meetings. Frame it positively: "AI Innovation Share." Ask team members to briefly share: "Here's a task where I used AI this week (e.g., summarizing a transcript, drafting initial discovery requests, brainstorming arguments), here's the tool/prompt I used, here's how it helped (saved time, improved quality, sparked an idea)," OR "Here's a situation where AI gave me a flawed output (e.g., missed a key exception, cited an overturned case), here's how I caught it, and what I learned about prompting or verification."
Internal Forums/Channels: Set up a dedicated internal chat channel (e.g., #AI_Practice_Tips) or a shared document where lawyers can quickly post useful prompts, links to helpful articles, ask questions about specific tools ("Has anyone tried using Tool X for contract review?"), or flag potential pitfalls they've encountered. Encourage interaction.
"AI Champions" & Mentorship: Identify individuals within different practice groups or seniority levels who are early adopters and naturally curious. Empower them as "AI Champions" – not necessarily experts, but go-to people for basic questions, who can help onboard colleagues and share best practices specific to that group's work. Consider reverse mentorship opportunities where junior associates familiar with the tech can help senior lawyers.
Lunch & Learns / Workshops: Host informal sessions focused on specific AI applications (e.g., "Using AI for Deposition Prep," "Ethical AI Use Cases in Litigation") or tools, encouraging open discussion and hands-on practice in a low-stakes environment.
Acknowledge the Learning Curve: Leadership must explicitly state that experimentation (within ethical bounds and firm guidelines) is encouraged and that making mistakes while learning is acceptable, provided they are caught and learned from. Fear of looking foolish or making an error is a major barrier to adoption.
The key is making conversation about AI use – including the mistakes and the learning process – a normal, expected part of the workflow. When people see colleagues successfully leveraging these tools and openly discussing how they caught errors or refined their approach, it demystifies the technology, reduces fear, accelerates collective learning, and builds trust. This cultural shift, supported by structural opportunities for sharing, is essential for moving beyond pockets of innovation to firm-wide adaptation. Overcoming resistance, especially from those comfortable with established methods, requires demonstrating value, providing support, and fostering this open dialogue.
Remaining Slides (blame substack for limits on the amount of content per post! You can see the slides here).
SLIDE 7
Now, let's revisit training. The traditional model often involves a "one-and-done" session: "Here's our new AI tool, here are the basic features, good luck." This approach is fundamentally flawed and frankly, irresponsible in the rapidly evolving AI landscape. Training focused solely on today's tools or specific skills like "prompt engineering" will quickly become obsolete. Prompt engineering, while useful now, might be less critical as interfaces become more intuitive or models better understand natural language requests through techniques like retrieval-augmented generation (RAG) which pull directly from specified document sets.
The goal shouldn't be just to teach people how to use the current tool; it must be to cultivate deep and enduring AI literacy. We need to equip our lawyers with the foundational understanding and critical thinking skills to evaluate, adapt to, and effectively utilize whatever AI tools emerge next month or next year.
SLIDE 8
Think of the proverb: "Give a man a fish, and you feed him for a day. Teach a man to fish, and you feed him for a lifetime." We need to go a step further. It's not enough to just teach them how to fish (i.e., use today's AI tool). We need to teach them to understand the fish – understand the underlying principles of how these models work (at a conceptual level – pattern recognition, probability, not true understanding), their strengths (speed, breadth, pattern matching), their inherent limitations (bias from training data, potential for fabrication, lack of common sense, difficulty with novel situations), the critical importance of verification against primary sources, and the non-negotiable ethical guardrails.
This "understanding the fish" approach allows lawyers to:
Critically evaluate outputs: Ask "Does this make sense?" not just "Is this formatted correctly?" Recognize when an AI might be confidently wrong. Share Michael Jordan example — just yesterday I interviewed a data scientist at Anthropic who penned one of the leading AI interpretability reports. When models think they know something, they override safeguards intended for them to be less confident.
Adapt strategies: Understand why a certain prompt works better than another for a specific task and adjust as models change. If a new tool emerges, they can assess its potential and limitations based on its underlying technology type.
Identify appropriate use cases: Know when AI is a powerful assistant (e.g., first drafts, summaries, brainstorming) versus when it's inappropriate or risky (e.g., final legal advice without review, complex strategic decisions).
Develop novel applications: See opportunities to use AI creatively for tasks we haven't even considered yet, because they understand the capabilities, not just the current features.
So, yes, provide initial, practical training on the specific tools your firm adopts. Use AI for initial feedback loops – have juniors prompt AI for partner-level reviews of their work, focusing on clarity, structure, and identifying potential gaps. But ensure your ongoing training program focuses on:
Core Concepts: How LLMs work (simplified), common pitfalls (hallucinations, bias), data privacy/security implications of different AI architectures (e.g., closed vs. open models, API use).
Critical Evaluation: Techniques for verifying AI output (checking citations, cross-referencing with trusted sources, logical consistency checks), spotting bias, assessing source reliability.
Ethical Use: Deep dives into duties of confidentiality (avoiding inputting sensitive client data), competence (understanding tool limits), supervision (oversight of AI and juniors using AI), and candor to the tribunal.
Adaptive Learning: Workshops on evaluating new AI tools, case studies of successful (and unsuccessful) AI integration, encouraging experimentation within safe boundaries and continuous learning about new developments through curated resources.
Training must be continuous, iterative, embedded in the workflow, and focused on building lasting literacy, not just temporary tool proficiency. It's an ongoing investment in your firm's future adaptability.
SLIDE 9
Take a minute to summarize this recent opinion from the Ohio Supreme Court. And, yes, I mean one minute.
SLIDE 10
I’ll give Gemini ten seconds to do the same.
Slide 11
As we can see, legal research is another area ripe for transformation, moving beyond simple keyword searches. AI should become the first stop for many research tasks, acting as an incredibly efficient research assistant. As you can see from this excerpt, it’s an incredibly efficient way to dive into an opinion.
Do I think this is perfect? No! But I’ll again emphasize that this is the worst AI you’ll ever use. I’ll also state that even with such improvements there are a couple best practices to adhere to:
Next Steps (Crucial):
Verification: Take the cases and conclusions cited by the AI and pull them up in Westlaw/Lexis. Confirm they are still good law. Read the relevant sections yourself. Did the AI accurately summarize the holding? Did it miss crucial nuance?
Deep Dive: Use the AI's output and your initial verification to formulate more targeted searches in traditional databases. Look for recent unpublished decisions, specific factual scenarios, or dissenting opinions the AI might have missed.
Analysis & Application: That’s still up to you - the creative part of lawyering, the judgment, the wisdom, that’s where your years of experience come in.
In short, AI isn't the end of the research process, but it's a dramatically more efficient start. It allows lawyers to quickly grasp the landscape, understand the key legal framework, formulate precise search queries, and identify the most relevant avenues for deeper investigation.
Mandating AI as the starting point for background research and initial issue spotting saves significant time and focuses expensive human effort on verification, analysis, and strategic application.
SLIDE 12
Let's talk about the elephant in the room: the billable hour. AI's ability to perform tasks faster and more efficiently poses a direct challenge to traditional billing models, and clients are becoming increasingly aware of these capabilities. Firms that cling rigidly to billing purely by time for tasks that AI can significantly accelerate – like initial document review, drafting standard clauses, or basic legal research – may find themselves struggling to justify their fees compared to more forward-thinking competitors. Ignoring this is perilous.
This isn't necessarily a threat; it's an opportunity to align your billing practices with the value you deliver. AI enables firms to explore and embrace Alternative Fee Arrangements with greater confidence and profitability.
Fixed Fees: When AI can reliably handle significant portions of document review for due diligence (reducing review time by 50-70% in some cases) or draft standard contracts much faster, offering fixed fees for these services becomes more predictable for the firm and highly appealing to clients seeking cost certainty.
Subscription Models: For ongoing compliance monitoring, routine corporate governance, or specific types of advisory work, firms could offer subscription services, leveraging AI to manage information flow, generate initial alerts or drafts, and allow lawyers to provide focused oversight and advice efficiently.
Value Billing / Success Fees: Focusing on the value delivered (e.g., a percentage of deal value, a success fee in litigation) rather than the hours spent becomes more viable when AI handles the commodity work, allowing lawyers to concentrate on high-value strategic advice, negotiation, and complex problem-solving where their judgment is paramount.
Hybrid Models: Combining hourly rates for bespoke strategic work with fixed fees or capped fees for more predictable, AI-assisted tasks.
Crucially, this requires proactive client communication. Don't wait for clients to ask why a task took X hours when they know AI exists. Explain how you are using AI to improve efficiency and quality, and how your billing reflects that value. Frame it as a benefit: "We leverage AI for initial document review to reduce costs and allow our team to focus on identifying the key strategic issues for you." Firms that lean into these new models, communicate transparently, and leverage AI to enhance efficiency while offering predictable pricing won't just attract more clients – they'll build stronger, trust-based relationships and likely retain them by demonstrating clear value and innovation.
SLIDE 13
All these changes point to a fundamental shift in the lawyer's role. If AI can draft the first version of the brief, conduct the initial background research, analyze thousands of documents for key clauses overnight, and even predict potential arguments, what's left for us?
The answer is: the most important, most human parts.
Our primary value proposition is evolving from doing the work (the writing, the researching) to exercising judgment about the work.
Judgment about what needs to be researched in depth, beyond the AI's initial pass.
Judgment about which arguments are most persuasive in this specific context, before this specific judge, against this specific opponent.
Judgment about what risks are acceptable to this particular client, considering their business goals, risk tolerance, and financial situation.
Judgment about when to settle and when to fight, based on a holistic assessment AI cannot make.
Judgment about what ultimately gets filed with the court or sent to opposing counsel – ensuring accuracy, tone, and strategic alignment.
Strategic Oversight: Designing the overall case strategy, identifying the key leverage points, and deciding how to deploy resources (including AI tools) most effectively.
AI can generate options, analyze data, and execute tasks with incredible speed, but it cannot (yet) replicate strategic foresight, ethical reasoning, creative problem-solving, or nuanced understanding of a client's business, industry, and personal context.
And critically, AI cannot replicate human connection. As former Ohio State Bar Association President Randall Comer aptly pointed out, "Artificial intelligence may already be able to dispense algorithmic, clinical legal advice. However, AI cannot provide clients with what they need most when confronted with legal problems—compassion."
In an age where information is abundant and tasks are automated, our ability to listen actively, understand unspoken concerns, empathize with distress, build trust, and guide clients through stressful, complex situations becomes not just valuable, but our core differentiator. Our role shifts definitively towards being trusted advisors, strategic thinkers, and compassionate counselors, leveraging AI as a powerful tool to enhance, not replace, our ability to deliver that uniquely human value.
SLIDE 14
Finally, let's be clear: embracing AI isn't just a business strategy; it aligns directly with, and is increasingly demanded by, our ethical obligations as lawyers here in Ohio.
The Ohio Rules of Professional Conduct are not static; they evolve with the practice. Consider Rule 1.1, requiring competence. Comment 8 explicitly states: "To maintain the requisite knowledge and skill, a lawyer should keep abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology..." Ignoring AI, failing to understand its capabilities, limitations, potential biases, and ethical implications, is arguably falling short of this duty of technological competence.
SLIDE 15
The AI imperative is clear. The technology will only improve, becoming more capable and integrated. The friction against its use will lessen as infrastructure and norms adapt. The firms that thrive – not just survive, but lead – will be those that move beyond mere tool adoption to systemic and cultural integration. This means rethinking hiring, actively fostering knowledge sharing and a culture of safe experimentation, committing to continuous AI literacy training, courageously adapting billing models, and fundamentally redefining the core value proposition of the lawyer.
This isn't about replacing lawyers; it's about augmenting them, freeing us to focus on our highest value: judgment, strategy, critical thinking, and compassion. It's about fulfilling our ethical duty to remain competent, protect our clients, and improve the delivery of legal services through ongoing adaptation and learning.
The time to prepare is not when AI is perfected. It's not next year. It's now. Start the conversations within your firms today. Build the structures for sharing and learning. Pilot AI tools thoughtfully. Begin revising your processes. Embrace the imperative to adapt, innovate, and lead the legal profession into its next, inevitable era.
Thank you.