Faegre Drinker formally deployed Harvey and Microsoft Copilot across the firm in June 2026, while Indiana-linked peers including Ice Miller and Barnes & Thornburg are testing or scaling their own legal AI stacks as clients push for faster, cheaper, and more transparent legal work. The Indiana Lawyer’s reporting captures a market that has moved beyond curiosity and into procurement, training, governance, and billing pressure. The real story is not that lawyers are “using AI.” It is that firms are beginning to decide which parts of legal work belong to lawyers, which parts belong to software, and which parts clients will no longer tolerate paying for at old rates.
For years, the legal industry treated artificial intelligence as a conference topic: a thing to be discussed on panels, demoed in innovation labs, and filed away as something that might matter once courts, clients, and malpractice insurers caught up. That phase is ending. Faegre Drinker’s firmwide rollout of Harvey and Microsoft Copilot, first announced by the firm on June 11 and examined locally by The Indiana Lawyer, is a useful marker because it shows AI moving from pilot project to infrastructure.
The firm is not alone. Barnes & Thornburg says nearly 90 percent of its attorneys are actively using AI, with more than 1,000 users logging into Harvey since deployment and more than 150,000 prompts submitted across its AI platforms in the last 30 days, according to the firm’s July announcement reported by Global Legal Chronicle. Ice Miller, meanwhile, is still in the evaluation stage, testing Harvey after recently trying Legora. That spread of approaches tells us more than any single adoption announcement could: the market has not settled, but the question has changed from whether to use AI to which mix of tools deserves institutional trust.
That distinction matters because legal work is unusually resistant to sloppy automation. A bad draft is not merely embarrassing; it can waive privilege, misstate law, misread a contract, or create downstream liability. Yet the work is also full of repetitive, language-heavy tasks that look almost purpose-built for generative AI: summarizing large document sets, drafting first-pass clauses, comparing versions, extracting obligations, preparing timelines, and organizing due diligence.
The pressure is therefore coming from both sides. Vendors are promising speed and scale, while clients are increasingly asking why they should pay premium hourly rates for work that software can accelerate. Firms that pretend nothing has changed risk looking inefficient. Firms that rush too far risk converting professional judgment into a user interface.
That is the right conceptual division. Microsoft Copilot is strongest where the work lives inside Outlook, Word, Excel, Teams, and the broader Microsoft 365 estate. It can summarize meetings, help draft correspondence, extract points from documents, and assist staff and professionals whose work is not exclusively legal analysis. In a law firm, that matters because not every AI use case is a brief, a contract, or a memo. A great deal of firm productivity is administrative, operational, and collaborative.
Harvey, by contrast, is being sold into the premium legal workflow layer. The company describes itself as an AI platform for legal and professional services, and its pitch rests on domain-specific workflows, legal research integrations, document analysis, and custom agents. The company says it is used by more than 142,000 professionals and 1,500 law firms across 60 countries, with more than 25,000 custom agents operating across use cases such as M&A, due diligence, contract drafting, and document review.
The attraction is obvious. If Copilot is the AI assistant sitting inside the office suite, Harvey is trying to become the workbench where legal professionals ask higher-stakes questions of messy legal material. Faegre Drinker’s earlier internal sandbox, Atlas AI, helped acclimate attorneys to the concept, Angelo told The Indiana Lawyer, but the firm ultimately decided it needed more than a safe ChatGPT-like training ground. That is the arc many enterprises are following: internal experiments first, then governed access to commercial systems with deeper integrations and vendor support.
But the two-layer model also creates a management problem. Once firms give employees both a general-purpose AI system and a legal-specialist AI system, they have to teach users not only how to prompt but where to work. The wrong tool for the task can create unnecessary risk, duplicate effort, or produce answers that look polished without being anchored in the right legal materials.
That is not corporate window dressing. It is the emerging minimum viable governance model for generative AI in professional services. Lawyers have duties of competence, confidentiality, supervision, candor, and communication. AI touches all of them at once. If a junior lawyer uses a tool to summarize discovery materials, a partner must understand the limits of the output. If a litigation team feeds strategy into a system without proper controls, privilege and confidentiality questions follow. If a generated memo cites nonexistent authority, the court will not accept “the model said so” as a defense.
The industry has already seen enough sanctions and embarrassment from fabricated citations to understand the danger. The lesson is not that AI cannot be used in legal practice. The lesson is that AI must be used inside a workflow that assumes it can be wrong, incomplete, or overconfident. A lawyer who treats AI output as a final answer is not delegating; that lawyer is abdicating.
This is where the legal profession’s old habits may help. Law firms already understand conflicts checks, document retention, ethical walls, matter management, and supervised review. AI governance has to be inserted into those existing muscles rather than presented as a novelty from the innovation team. The firms that succeed will be the ones that make AI review feel like part of ordinary professional discipline, not a separate compliance theater.
Angelo’s language about “level-setting” is revealing. The purpose of training is not merely to reduce risk. It is to create a shared institutional vocabulary so that lawyers, paralegals, staff, technologists, and clients can discuss AI use without pretending everyone means the same thing. In that sense, governance is also a product strategy.
That number is not just a financing milestone. It is a market signal. Investors are betting that legal AI will not collapse into a generic chatbot feature inside Microsoft Word or Google Docs. They are betting that law firms and legal departments will pay premium prices for domain-specific systems that understand documents, matters, permissions, citations, workflows, and professional risk.
The bear case is equally clear. If the underlying models come from OpenAI, Anthropic, Google, and other foundation-model companies, how much defensible value sits in the legal application layer? Harvey says its platform routes tasks across multiple large language models and decomposes requests into subtasks to produce better outputs. That kind of orchestration can matter. But customers will eventually ask whether the difference is worth the price, especially as Microsoft, Thomson Reuters, LexisNexis, Relativity, Legora, CoCounsel, and others push deeper into the same territory.
The controversy around a now-deleted Reddit post reportedly written by a former Harvey employee shows the other side of the hype cycle. The anonymous post criticized adoption and cost, and it circulated widely in the LegalTech community before Harvey’s leadership pushed back publicly and shared metrics. Anonymous posts are not audited financials, and they should be treated cautiously. But the fact that the discussion resonated says something real about the market’s anxiety: legal AI vendors are being valued like category winners before buyers have fully figured out daily usage, renewal discipline, and measurable client value.
That tension is not unique to Harvey. It is the defining tension of enterprise AI in 2026. Usage can be broad but shallow. Pilots can be exciting but hard to convert into durable workflow change. A tool can save time on one task and create review burden on another. The winners will be judged less by demo quality than by whether lawyers keep using the software after the novelty fades.
The firm’s co-chair of its AI practice, Brian McGinnis, put the issue bluntly in the firm’s release: AI is not taking work away from lawyers; it is changing which parts of the work clients are willing to pay for. That may be the most honest sentence in the whole legal AI debate. The threat is not a robot lawyer replacing a partner in a boardroom. The threat is that clients start disaggregating legal labor and refusing to pay bespoke rates for commodity process.
Barnes & Thornburg’s reported usage numbers are striking. Nearly 90 percent active attorney usage is unusually high for a new enterprise tool, and 150,000 prompts in 30 days suggests AI is no longer confined to a small innovation cohort. The firm’s stack includes Harvey, CoCounsel, ChatBT, and practice-specific tools, which shows that even aggressive adopters are not betting on a single system.
That multi-tool reality is likely to persist. Different legal tasks reward different architectures. Litigation teams may care about discovery review, privilege logs, deposition preparation, and case-law analysis. Corporate teams may care about diligence, contract comparison, disclosure schedules, and negotiation playbooks. Labor, IP, tax, restructuring, and regulatory practices each have their own documents, rhythms, and risk tolerances.
The “practice champion” model acknowledges that reality. Central IT can vet security, negotiate contracts, and define policy, but it cannot discover every useful workflow inside every practice group. Lawyers who actually use the tools on live matters are better positioned to identify where the technology helps, where it fails, and where clients may accept or reject AI-assisted work.
That caution is not indecision. It may be prudence. The legal AI market is changing too quickly for every firm to standardize prematurely. Tools are improving, pricing models are unsettled, integrations are evolving, and client expectations are not uniform. A mid-sized or regional firm that signs a large, expensive, multi-year deployment too early could find itself locked into the wrong abstraction layer.
Boyer’s comment that the market may take four, five, or six years to settle is plausible. In fact, it may be optimistic. The legal industry is still working through basic questions. Should AI usage be disclosed to clients? When should clients receive discounts for AI-assisted work? Who owns prompts and outputs? How should firms preserve AI work product? Can prompts be discoverable? How should firms supervise nonlawyer use of AI tools? Which systems are acceptable for confidential information?
Those questions do not have one-size-fits-all answers. A public company merger, a criminal defense matter, a patent prosecution file, and a routine commercial contract review present different risks. The right AI strategy may therefore be less about choosing the platform and more about building a repeatable evaluation discipline.
Ice Miller’s approach reflects that. Test, compare, train, observe, and avoid treating vendor hype as a substitute for client value. In a market flooded with confident demos, skepticism is not backward-looking. It is operational hygiene.
For WindowsForum readers, this is the familiar enterprise pattern. The specialized vendor gets the headlines, while Microsoft tries to make the capability feel native, governed, and procurement-friendly. If Copilot can summarize Teams calls, draft emails, interrogate Word documents, analyze Excel data, and respect Microsoft 365 permissions, then it becomes the default AI surface for a huge amount of firm work that is not narrowly “legal AI.”
That has two consequences. First, law firms may end up with more AI usage in administrative and operational workflows than in formal legal analysis, at least at the beginning. Second, legal AI vendors must prove that their domain-specific capabilities justify a place alongside a Microsoft product that many firms are already licensing or evaluating.
This does not mean Copilot replaces Harvey. The better analogy is the relationship between Windows and specialized professional software. Microsoft supplies the broad platform; vertical applications survive by doing high-value domain work better than the platform can. The question is how much of today’s legal AI functionality remains vertical once general-purpose enterprise AI improves.
Law firms are, in effect, testing that boundary in real time. Faegre Drinker’s two-platform rollout says there is room for both. Barnes & Thornburg’s stack says there is room for more than two. Ice Miller’s ongoing testing says the boundary is still contested.
This is where legal AI collides with the billable hour. Many AI use cases are designed to reduce time spent on tasks. Traditional law firm economics monetize time. That does not make adoption impossible, but it does create tension. A tool that helps an associate review documents faster is valuable to the client, valuable to the lawyer, and potentially disruptive to a billing model built around hours.
The firms that handle this well will not merely say “we use AI responsibly.” They will develop pricing models, matter plans, and client communications that explain how AI changes the work. Some tasks may move to fixed fees. Some may become part of value-based billing. Some may remain hourly because the bottleneck is still expert judgment, not document handling.
Clients are also becoming more sophisticated about AI risk. They may ask which tools are approved, whether their data is used for model training, how outputs are reviewed, whether AI use is logged, and how privilege is protected. A firm that can answer those questions clearly will have an advantage over one that treats AI as an informal productivity hack.
That is why the Indiana examples matter beyond Indiana. They show firms trying to turn AI from an individual behavior into an institutional promise. The client does not want to hear that one associate found a useful prompt. The client wants to know that the firm has a defensible system.
First drafts will be easier to produce. Summaries will be faster. Contract comparisons will become more automated. Research paths will be generated in minutes. The premium will shift toward framing the right question, identifying what the model missed, validating the answer, advising the client, and taking responsibility for the result.
That is not a small change. Junior lawyers have traditionally learned by doing some of the very tasks AI is now poised to accelerate. Due diligence, document review, research memos, and first-pass drafting are not glamorous, but they are part of professional formation. If firms automate too much without redesigning training, they may solve a short-term efficiency problem by creating a long-term talent problem.
There is also a cultural risk. AI output often sounds competent even when it is thin. In law, style can camouflage weakness. A generated paragraph with the right tone, structure, and vocabulary may still miss the controlling authority, misunderstand a factual nuance, or overstate a conclusion. The danger is not that lawyers will trust obviously bad work. The danger is that they will trust plausible work because it resembles the form of competence.
This is why “human in the loop” cannot be a slogan. The human must be qualified, attentive, accountable, and given enough time to review. If AI simply compresses deadlines while leaving lawyers responsible for validating more machine-generated material in less time, the quality gains may prove illusory.
They must decide whether to buy from legal-specific startups, incumbent research providers, Microsoft, internal development teams, or all of the above. They must decide how to govern tools that evolve faster than traditional software. They must decide how much experimentation to allow at the practice level and how much standardization to impose from the center.
The old law-firm technology cycle moved slowly. Document management systems, timekeeping software, e-discovery platforms, and research tools were adopted over years. Generative AI is moving faster because individual lawyers can feel the utility immediately. Once someone uses AI to summarize a 200-page document set or produce a rough first draft in minutes, the old workflow starts to feel irrational.
But immediate utility is not the same as institutional readiness. Security reviews, data handling, retention rules, client consent, professional responsibility, and training all have to catch up. That is why the most interesting firms are not simply the ones with the highest prompt counts. They are the ones connecting usage to governance and client value.
Faegre Drinker, Barnes & Thornburg, and Ice Miller represent three points on that curve: enterprise rollout with mandatory training, high-adoption embedded practice champions, and careful comparative testing. None is obviously wrong. Each reflects a different answer to the same question: how much certainty does a firm need before it changes how legal work gets done?
Legal AI Has Left the Demo Room and Entered the Billing Conversation
For years, the legal industry treated artificial intelligence as a conference topic: a thing to be discussed on panels, demoed in innovation labs, and filed away as something that might matter once courts, clients, and malpractice insurers caught up. That phase is ending. Faegre Drinker’s firmwide rollout of Harvey and Microsoft Copilot, first announced by the firm on June 11 and examined locally by The Indiana Lawyer, is a useful marker because it shows AI moving from pilot project to infrastructure.The firm is not alone. Barnes & Thornburg says nearly 90 percent of its attorneys are actively using AI, with more than 1,000 users logging into Harvey since deployment and more than 150,000 prompts submitted across its AI platforms in the last 30 days, according to the firm’s July announcement reported by Global Legal Chronicle. Ice Miller, meanwhile, is still in the evaluation stage, testing Harvey after recently trying Legora. That spread of approaches tells us more than any single adoption announcement could: the market has not settled, but the question has changed from whether to use AI to which mix of tools deserves institutional trust.
That distinction matters because legal work is unusually resistant to sloppy automation. A bad draft is not merely embarrassing; it can waive privilege, misstate law, misread a contract, or create downstream liability. Yet the work is also full of repetitive, language-heavy tasks that look almost purpose-built for generative AI: summarizing large document sets, drafting first-pass clauses, comparing versions, extracting obligations, preparing timelines, and organizing due diligence.
The pressure is therefore coming from both sides. Vendors are promising speed and scale, while clients are increasingly asking why they should pay premium hourly rates for work that software can accelerate. Firms that pretend nothing has changed risk looking inefficient. Firms that rush too far risk converting professional judgment into a user interface.
Faegre Drinker Is Betting on a Two-Layer AI Stack
Faegre Drinker’s approach is notable because it does not frame Harvey and Copilot as interchangeable. Scott Angelo, the firm’s chief technology and innovation officer, told The Indiana Lawyer that the firm wanted to become an “AI-led organization” and put AI in the hands of everyone. But the tool split is deliberate: Copilot serves as the broad productivity layer across Microsoft 365, while Harvey is positioned as the legal-specialist environment for attorney workflows.That is the right conceptual division. Microsoft Copilot is strongest where the work lives inside Outlook, Word, Excel, Teams, and the broader Microsoft 365 estate. It can summarize meetings, help draft correspondence, extract points from documents, and assist staff and professionals whose work is not exclusively legal analysis. In a law firm, that matters because not every AI use case is a brief, a contract, or a memo. A great deal of firm productivity is administrative, operational, and collaborative.
Harvey, by contrast, is being sold into the premium legal workflow layer. The company describes itself as an AI platform for legal and professional services, and its pitch rests on domain-specific workflows, legal research integrations, document analysis, and custom agents. The company says it is used by more than 142,000 professionals and 1,500 law firms across 60 countries, with more than 25,000 custom agents operating across use cases such as M&A, due diligence, contract drafting, and document review.
The attraction is obvious. If Copilot is the AI assistant sitting inside the office suite, Harvey is trying to become the workbench where legal professionals ask higher-stakes questions of messy legal material. Faegre Drinker’s earlier internal sandbox, Atlas AI, helped acclimate attorneys to the concept, Angelo told The Indiana Lawyer, but the firm ultimately decided it needed more than a safe ChatGPT-like training ground. That is the arc many enterprises are following: internal experiments first, then governed access to commercial systems with deeper integrations and vendor support.
But the two-layer model also creates a management problem. Once firms give employees both a general-purpose AI system and a legal-specialist AI system, they have to teach users not only how to prompt but where to work. The wrong tool for the task can create unnecessary risk, duplicate effort, or produce answers that look polished without being anchored in the right legal materials.
Training Is Becoming the New Malpractice Firewall
Faegre Drinker’s most important AI decision may not be the vendor selection. It may be the gating process. Before employees get access, the firm requires basic AI training, professional responsibility training, and testing on a firm-specific AI security policy, according to The Indiana Lawyer.That is not corporate window dressing. It is the emerging minimum viable governance model for generative AI in professional services. Lawyers have duties of competence, confidentiality, supervision, candor, and communication. AI touches all of them at once. If a junior lawyer uses a tool to summarize discovery materials, a partner must understand the limits of the output. If a litigation team feeds strategy into a system without proper controls, privilege and confidentiality questions follow. If a generated memo cites nonexistent authority, the court will not accept “the model said so” as a defense.
The industry has already seen enough sanctions and embarrassment from fabricated citations to understand the danger. The lesson is not that AI cannot be used in legal practice. The lesson is that AI must be used inside a workflow that assumes it can be wrong, incomplete, or overconfident. A lawyer who treats AI output as a final answer is not delegating; that lawyer is abdicating.
This is where the legal profession’s old habits may help. Law firms already understand conflicts checks, document retention, ethical walls, matter management, and supervised review. AI governance has to be inserted into those existing muscles rather than presented as a novelty from the innovation team. The firms that succeed will be the ones that make AI review feel like part of ordinary professional discipline, not a separate compliance theater.
Angelo’s language about “level-setting” is revealing. The purpose of training is not merely to reduce risk. It is to create a shared institutional vocabulary so that lawyers, paralegals, staff, technologists, and clients can discuss AI use without pretending everyone means the same thing. In that sense, governance is also a product strategy.
Harvey’s Valuation Reflects Both Confidence and Anxiety
Harvey’s rise has been extraordinary. Founded in 2022 by Winston Weinberg and Gabriel Pereyra, the company has become one of the defining names in legal AI, helped by its relationship with OpenAI and by a broader investor belief that law is one of the most lucrative verticals for generative AI. In March 2026, Harvey confirmed a $200 million funding round co-led by GIC and Sequoia at an $11 billion valuation, as reported by TechCrunch and announced by the company.That number is not just a financing milestone. It is a market signal. Investors are betting that legal AI will not collapse into a generic chatbot feature inside Microsoft Word or Google Docs. They are betting that law firms and legal departments will pay premium prices for domain-specific systems that understand documents, matters, permissions, citations, workflows, and professional risk.
The bear case is equally clear. If the underlying models come from OpenAI, Anthropic, Google, and other foundation-model companies, how much defensible value sits in the legal application layer? Harvey says its platform routes tasks across multiple large language models and decomposes requests into subtasks to produce better outputs. That kind of orchestration can matter. But customers will eventually ask whether the difference is worth the price, especially as Microsoft, Thomson Reuters, LexisNexis, Relativity, Legora, CoCounsel, and others push deeper into the same territory.
The controversy around a now-deleted Reddit post reportedly written by a former Harvey employee shows the other side of the hype cycle. The anonymous post criticized adoption and cost, and it circulated widely in the LegalTech community before Harvey’s leadership pushed back publicly and shared metrics. Anonymous posts are not audited financials, and they should be treated cautiously. But the fact that the discussion resonated says something real about the market’s anxiety: legal AI vendors are being valued like category winners before buyers have fully figured out daily usage, renewal discipline, and measurable client value.
That tension is not unique to Harvey. It is the defining tension of enterprise AI in 2026. Usage can be broad but shallow. Pilots can be exciting but hard to convert into durable workflow change. A tool can save time on one task and create review burden on another. The winners will be judged less by demo quality than by whether lawyers keep using the software after the novelty fades.
Barnes & Thornburg Is Turning Adoption Into an Operating Model
Barnes & Thornburg’s latest move is important because it treats AI adoption as a people problem, not just a licensing problem. The firm says it now has 38 attorneys across 26 markets serving as AI practice champions, with those lawyers working inside their practices to advance AI use in client service. That is a more mature approach than simply buying seats and waiting for utilization charts to rise.The firm’s co-chair of its AI practice, Brian McGinnis, put the issue bluntly in the firm’s release: AI is not taking work away from lawyers; it is changing which parts of the work clients are willing to pay for. That may be the most honest sentence in the whole legal AI debate. The threat is not a robot lawyer replacing a partner in a boardroom. The threat is that clients start disaggregating legal labor and refusing to pay bespoke rates for commodity process.
Barnes & Thornburg’s reported usage numbers are striking. Nearly 90 percent active attorney usage is unusually high for a new enterprise tool, and 150,000 prompts in 30 days suggests AI is no longer confined to a small innovation cohort. The firm’s stack includes Harvey, CoCounsel, ChatBT, and practice-specific tools, which shows that even aggressive adopters are not betting on a single system.
That multi-tool reality is likely to persist. Different legal tasks reward different architectures. Litigation teams may care about discovery review, privilege logs, deposition preparation, and case-law analysis. Corporate teams may care about diligence, contract comparison, disclosure schedules, and negotiation playbooks. Labor, IP, tax, restructuring, and regulatory practices each have their own documents, rhythms, and risk tolerances.
The “practice champion” model acknowledges that reality. Central IT can vet security, negotiate contracts, and define policy, but it cannot discover every useful workflow inside every practice group. Lawyers who actually use the tools on live matters are better positioned to identify where the technology helps, where it fails, and where clients may accept or reject AI-assisted work.
Ice Miller’s Caution May Age Better Than It Looks
Against the backdrop of firmwide rollouts, Ice Miller’s posture may look conservative. James Boyer, the firm’s chief innovation officer, told The Indiana Lawyer that the firm has tested Legora and is now testing Harvey with some attorneys. He framed the decision around client needs: responsiveness, efficiency, and availability.That caution is not indecision. It may be prudence. The legal AI market is changing too quickly for every firm to standardize prematurely. Tools are improving, pricing models are unsettled, integrations are evolving, and client expectations are not uniform. A mid-sized or regional firm that signs a large, expensive, multi-year deployment too early could find itself locked into the wrong abstraction layer.
Boyer’s comment that the market may take four, five, or six years to settle is plausible. In fact, it may be optimistic. The legal industry is still working through basic questions. Should AI usage be disclosed to clients? When should clients receive discounts for AI-assisted work? Who owns prompts and outputs? How should firms preserve AI work product? Can prompts be discoverable? How should firms supervise nonlawyer use of AI tools? Which systems are acceptable for confidential information?
Those questions do not have one-size-fits-all answers. A public company merger, a criminal defense matter, a patent prosecution file, and a routine commercial contract review present different risks. The right AI strategy may therefore be less about choosing the platform and more about building a repeatable evaluation discipline.
Ice Miller’s approach reflects that. Test, compare, train, observe, and avoid treating vendor hype as a substitute for client value. In a market flooded with confident demos, skepticism is not backward-looking. It is operational hygiene.
The Microsoft Layer Is the Quiet Strategic Center
Because Harvey is the glamorous legal AI startup with the huge valuation, it naturally attracts attention. But Microsoft Copilot may be the more structurally important piece of the law-firm AI puzzle. Most firms already live inside Microsoft’s productivity stack, and that makes Copilot less like a new application and more like an ambient layer across daily work.For WindowsForum readers, this is the familiar enterprise pattern. The specialized vendor gets the headlines, while Microsoft tries to make the capability feel native, governed, and procurement-friendly. If Copilot can summarize Teams calls, draft emails, interrogate Word documents, analyze Excel data, and respect Microsoft 365 permissions, then it becomes the default AI surface for a huge amount of firm work that is not narrowly “legal AI.”
That has two consequences. First, law firms may end up with more AI usage in administrative and operational workflows than in formal legal analysis, at least at the beginning. Second, legal AI vendors must prove that their domain-specific capabilities justify a place alongside a Microsoft product that many firms are already licensing or evaluating.
This does not mean Copilot replaces Harvey. The better analogy is the relationship between Windows and specialized professional software. Microsoft supplies the broad platform; vertical applications survive by doing high-value domain work better than the platform can. The question is how much of today’s legal AI functionality remains vertical once general-purpose enterprise AI improves.
Law firms are, in effect, testing that boundary in real time. Faegre Drinker’s two-platform rollout says there is room for both. Barnes & Thornburg’s stack says there is room for more than two. Ice Miller’s ongoing testing says the boundary is still contested.
Clients Will Decide Whether AI Is Innovation or Margin Protection
Law firms like to describe AI as a way to serve clients better. That is true, but incomplete. Clients will judge AI adoption through a colder lens: does it reduce cost, improve speed, increase consistency, or produce better outcomes? If the answer is yes but the invoice looks the same, the conversation will get uncomfortable.This is where legal AI collides with the billable hour. Many AI use cases are designed to reduce time spent on tasks. Traditional law firm economics monetize time. That does not make adoption impossible, but it does create tension. A tool that helps an associate review documents faster is valuable to the client, valuable to the lawyer, and potentially disruptive to a billing model built around hours.
The firms that handle this well will not merely say “we use AI responsibly.” They will develop pricing models, matter plans, and client communications that explain how AI changes the work. Some tasks may move to fixed fees. Some may become part of value-based billing. Some may remain hourly because the bottleneck is still expert judgment, not document handling.
Clients are also becoming more sophisticated about AI risk. They may ask which tools are approved, whether their data is used for model training, how outputs are reviewed, whether AI use is logged, and how privilege is protected. A firm that can answer those questions clearly will have an advantage over one that treats AI as an informal productivity hack.
That is why the Indiana examples matter beyond Indiana. They show firms trying to turn AI from an individual behavior into an institutional promise. The client does not want to hear that one associate found a useful prompt. The client wants to know that the firm has a defensible system.
The New Legal Skill Is Knowing When Not to Automate
The most overheated version of the AI debate imagines a near-term collapse of legal labor. The more realistic version is subtler and more disruptive. AI will not eliminate the need for lawyers, but it will expose which pieces of legal work were never really about legal judgment.First drafts will be easier to produce. Summaries will be faster. Contract comparisons will become more automated. Research paths will be generated in minutes. The premium will shift toward framing the right question, identifying what the model missed, validating the answer, advising the client, and taking responsibility for the result.
That is not a small change. Junior lawyers have traditionally learned by doing some of the very tasks AI is now poised to accelerate. Due diligence, document review, research memos, and first-pass drafting are not glamorous, but they are part of professional formation. If firms automate too much without redesigning training, they may solve a short-term efficiency problem by creating a long-term talent problem.
There is also a cultural risk. AI output often sounds competent even when it is thin. In law, style can camouflage weakness. A generated paragraph with the right tone, structure, and vocabulary may still miss the controlling authority, misunderstand a factual nuance, or overstate a conclusion. The danger is not that lawyers will trust obviously bad work. The danger is that they will trust plausible work because it resembles the form of competence.
This is why “human in the loop” cannot be a slogan. The human must be qualified, attentive, accountable, and given enough time to review. If AI simply compresses deadlines while leaving lawyers responsible for validating more machine-generated material in less time, the quality gains may prove illusory.
The Indiana Story Is Really a National Procurement Story
It would be easy to read the Indiana angle as a local business story: Faegre Drinker expands AI use, Ice Miller tests tools, Barnes & Thornburg appoints champions. But the deeper pattern is national. Large and mid-sized firms are building AI procurement strategies under conditions of uncertainty.They must decide whether to buy from legal-specific startups, incumbent research providers, Microsoft, internal development teams, or all of the above. They must decide how to govern tools that evolve faster than traditional software. They must decide how much experimentation to allow at the practice level and how much standardization to impose from the center.
The old law-firm technology cycle moved slowly. Document management systems, timekeeping software, e-discovery platforms, and research tools were adopted over years. Generative AI is moving faster because individual lawyers can feel the utility immediately. Once someone uses AI to summarize a 200-page document set or produce a rough first draft in minutes, the old workflow starts to feel irrational.
But immediate utility is not the same as institutional readiness. Security reviews, data handling, retention rules, client consent, professional responsibility, and training all have to catch up. That is why the most interesting firms are not simply the ones with the highest prompt counts. They are the ones connecting usage to governance and client value.
Faegre Drinker, Barnes & Thornburg, and Ice Miller represent three points on that curve: enterprise rollout with mandatory training, high-adoption embedded practice champions, and careful comparative testing. None is obviously wrong. Each reflects a different answer to the same question: how much certainty does a firm need before it changes how legal work gets done?
The Firms That Win Will Treat AI as Practice Infrastructure
The concrete lessons from this moment are less flashy than the vendor announcements, but they are more durable. Legal AI is becoming infrastructure, and infrastructure has to be managed.- Firms are moving from AI pilots to firmwide deployments, but the most credible rollouts pair access with mandatory training, ethics guidance, and security rules.
- Harvey’s rapid growth and $11 billion valuation show investor confidence in vertical legal AI, but customers will still demand proof that usage translates into measurable value.
- Microsoft Copilot gives firms a broad productivity layer, while legal-specific tools such as Harvey, Legora, and CoCounsel compete for higher-risk domain workflows.
- Practice-level champions may be more effective than top-down mandates because legal AI use cases vary sharply across litigation, transactions, regulatory work, and specialty practices.
- Clients are likely to push hardest not on whether firms use AI, but on whether AI changes cost, speed, transparency, and the parts of legal work they are willing to pay for.
- The biggest long-term risk is not lawyer replacement; it is the erosion of training, review discipline, and professional judgment if firms automate work faster than they redesign supervision.
References
- Primary source: The Indiana Lawyer
Published: 2026-07-03T18:50:15.556512
- Related coverage: gic.com.sg
</rdf:Alt> </dc:title> <dc:description> <rdf:Alt> <rdf:li xml:lang="x-default"/> </rdf:Alt> </dc:description> <dc:creator> <rdf:Seq> <rdf:li>Julia T
</rdf:Alt> </dc:description> <dc:creator> <rdf:Seq> <rdf:li>Julia Tarantinowww.gic.com.sg
- Related coverage: insidelegalai.com
BonelliErede Rolls Out Harvey Firmwide After AI Task Force Review — Inside Legal AI
InsidePractice · The Next Frontier - original reporting on legal engineers, vibecoding, AI-native firms, new law models, agentic AI, and legal operations.www.insidelegalai.com - Related coverage: faegredrinker.com
faegre drinker partners with orgaimi leveraging ai to better anticipate client needs and strengthen firm relationships
PDF documentwww.faegredrinker.com
- Related coverage: btlaw.com
Barnes & Thornburg Ranks Among Leading Law Firms in 2026 Chambers USA | Barnes & Thornburg
Barnes &amp; Thornburg has once again been recognized as one of the top law firms in the nation in the 2026 edition of Chambers USA. The firm earned 49 practice
btlaw.com
