Akerman Hires a Litigator to Build Firm AI

Akerman named Charles Zerner its first director of AI development on Sept. 9, 2026, a job created to write legal software the vendors do not sell. The Am Law 100 firm says he will sit with lawyers and test tools such as Harvey, CoCounsel, and Microsoft Copilot. He will build custom apps inside Akerman’s private cloud when those products fall short.

The hire closes a three-seat bench that treats buy versus build as three jobs, not one shopping trip. Zerner is the build seat, and he reports into information technology rather than a practice group.

Akerman Split AI Into Three Jobs

Akerman’s announcement puts Zerner in a newly created Director of AI Development role on the IT team under Chief Information Officer Azfar Shamsi. Chairman and CEO Scott Meyers said the firm is running a hybrid strategy that decides when to buy, when to build, and when to combine the two. He called Zerner the person who can create a solution when the market does not offer what lawyers and clients need.

That sentence is the job, not a slogan. A June job posting for a builder said the opening was not a strategy or advisory seat. It asked for someone who had shipped production AI, including agentic systems, could read and write code, and could run those systems on confidential, privileged data.

AKERMAN’S 2026 AI BENCH

1. January 13, 2026: Launches a Law+AI Initiative with USC Gould School of Law.
2. May 2026: Centers a firmwide retreat of about 700 people on AI and credits billable hours for AI training.
3. June 3, 2026: Posts the Director of Artificial Intelligence opening that asks for a hands-on builder.
4. August 6, 2026: Hires Michael Adler as director of AI governance and data protection.
5. September 9, 2026: Names Charles Zerner director of AI development.

The sequence is the design. Adler arrived a month before Zerner to write the rules. Tricia Thomas already held the research seat. Zerner is the missing builder, hired after the firm spent the year teaching lawyers to use the tools and after it posted a job that refused a pure strategist.

AKERMAN’S THREE AI SEATS

• Research: Tricia Thomas, director of AI and research services, maps what commercial products already do.
• Development: Charles Zerner builds custom apps in the firm’s private cloud and data centers when those products fail a real workflow.
• Governance: Michael Adler, director of AI governance and data protection, sets how anything that ships is overseen.

Shamsi stated the split in one line, and it is the operating model hiding under the hire.

Research identifies what exists; development builds what doesn’t; and governance ensures we deploy it responsibly.

Azfar Shamsi, Chief Information Officer, Akerman

Plenty of firms now have an innovation partner and a stack of vendor licenses. Fewer have split “what should we do with AI” into evaluate, make, and police, then staffed each seat with a different specialist. Zerner’s arrival is the make seat going live.

The Litigator Who Writes Production Code

Zerner joins from Munck Wilson Mandala in Dallas, where he was a principal attorney in complex commercial litigation and sat in the firm’s AI and machine learning practice. Akerman said he has more than 13 years of high-stakes, defense-side work and more than 25 years with artificial intelligence. He earned a B.A. with honors from Harvard and a J.D., summa cum laude, from Tulane University Law School, where he was articles editor of the Tulane Law Review and elected to the Order of the Coif.

He co-founded a neural-network startup in 2000, while still an undergraduate, work that later profiles have tied to cancer-treatment research. He has used machine learning in legal practice since 2012, the year he finished law school, and has taught continuing legal education on AI since 2017, including ethics programs for the Texas Center for Legal Ethics. Munck Wilson’s biography put his commercial disputes in a range from $50,000 to more than $1 billion.

The file Akerman chose to lead with is not a paper. The firm said Zerner designed and built a document-review platform on open-source models, running entirely inside Munck Wilson’s control, that automated review of more than 100 million documents in a major intellectual property dispute and saved the client millions. Shamsi called him a Harvard-educated litigator who writes production code, tunes open-weight models, and builds applications that run inside a controlled environment.

That is why the seat sits in IT. Adler joined as a partner in the corporate group as well as a governance director. Zerner’s mandate is to ship software. He will work directly with attorneys and practice groups, decide whether Harvey, CoCounsel, or Copilot can handle a workflow, and write a bespoke application when they cannot. The person translating the legal problem is supposed to be the person who types the fix.

Akerman also said he was chief legal architect and primary drafter in Vacker et al. v. ElevenLabs, which the firm described as the first U.S. generative-AI copyright case to reach a settlement. The dual track is the point of the hire. He has billed like the users of these tools, and he has already built one of them at production scale.

What $500 Million Buys at Kirkland

Akerman is not placing this bet in a vacuum. Kirkland & Ellis, the highest-grossing firm in the world, set aside $500 million to build its own AI platform, with chair Jon Ballis saying the firm expects to spend more than $100 million in 2026 and hundreds of millions more over the next three to four years. Kirkland reported $10.6 billion in revenue last year. About 180 technology professionals are on the work, and about 250 lawyers, including about 100 partners, are feeding it how they actually practice. The firm also launched a Palantir relationship aimed first at private-equity fundraising.

Ballis’s case for that check was blunt. Widely available tools, he said, raise the floor for everyone, and Kirkland does not get hired for the floor. The spend comes out of revenue, which cuts what equity partners take home in the near term. It is a wager that owning the stack is worth a smaller distribution now.

HOW FIVE FIRMS ARE PLACING THE AI BET

Gretta Rusanow, managing director and head of advisory services at Citigroup’s law firm group, has said industry spending on AI is set to climb through 2026. The table is the split that climb is producing. Some firms are writing nine-figure checks. Some are buying chips. Some are hiring one person who can walk away from a vendor. Some are still renting.

Latham’s Nvidia Cluster and Client Data

Latham & Watkins, which reported $8.3 billion in revenue last year, confirmed it has bought Nvidia GPU servers and begun customizing models on hardware only its people can touch. Leaders of the firm’s AI strategy committee said the server plan started three years ago. A spokesperson said the firm has purchased multiple H200 GPUs and is looking at newer Nvidia systems. The boxes sit in a third-party data center that Latham staff alone can enter. Engineers are fine-tuning Nvidia’s Nemotron 3 open-weight models, which can be downloaded and run without sending every token to a closed model vendor.

CIO Rene Mendoza put the reason in client terms, not lab terms.

Sometimes we may have information that is so sensitive, client information that we really want to protect, we don’t want to put it to any cloud vendor.

Rene Mendoza, Chief Information Officer, Latham & Watkins

He also pointed to consumption pricing. As usage bills rise, owning the computers is a hedge as well as a confidentiality wall. Michael Rubin, global chair of Latham’s AI practice and chair of its internal AI strategy committee, said integrating open-source models with the firm’s own software on an on-premises basis puts Latham in a class other firms cannot match. The trade is capital and a security team. The firm now owns uptime, patches, and the blast radius if something goes wrong.

Among lawyers who left practice to sell tools, the server news landed as a status marker. The jokes assumed the cluster would work well enough to cause new problems, not that a law firm had no business touching GPUs. That is a change from the years when in-house legal AI meant a branded chatbot on a vendor’s cloud.

Why Most Firms Still Rent Their Tools

Ropes & Gray took the other fork. Stein, the CIO, said the technology is changing so quickly that a proprietary build does not make sense, and that buying is the better option for now. Jane Rogers, a member of the firm’s management committee, has said the firm spends a lot of time on whether to build an independent AI platform or keep a more open setup. For anyone outside the very largest shops, the choice is often not real. Offit Kurman chief innovation officer Alex Finkel has said that beyond the top 50 firms, the build-versus-buy selection is not applicable.

K&L Gates sat in the middle, and its reason is the one Zerner’s job is built to catch. The firm said it built an internal tool to draft SEC Form N-14 submissions after testing third-party software that could not handle the form’s requirements. In testing, the tool cut the billable hours for that filing in half. That is not a platform strategy. It is a single painful workflow that the market did not productize, which is the exact test Shamsi described.

Building still has a long tail. A homegrown tool has to be patched, evaluated, and retrained after the people who loved the prototype have gone back to billable work. Buying moves that burden onto a vendor whose whole business is to carry it. The cost of being wrong is not the license fee. It is a half-finished internal app that only one associate can run, sitting next to a vendor product the firm already pays for and does not fully use.

Harvey, Copilot, and the Shared Floor

Akerman did not pretend those products are going away. The press release names Harvey, CoCounsel, and Microsoft Copilot as the commercial layer Zerner will test first. That is the floor Ballis talked about, the shared kit every well-resourced firm can rent. The research seat exists because that kit keeps moving. The governance seat exists because agentic systems do not always do what they were told. Meyers has spent 2026 writing about that gap, including a July note on how enterprise rules built for obedient tools fail when the software starts making plans.

Partnership talk has filled the same week as these hires and hardware buys. Firms want early access and a say in the roadmap, and vendors want data and distribution. A partnership is still a vendor relationship if the firm cannot reproduce the work when the contract sours. The walk-away option is an engineer who already knows the matter type, the privilege problem, and the model’s failure mode.

Zerner’s 100 million-document system is the specimen. It ran on open-source models inside the firm’s own walls. It was aimed at a review pile large enough that a generic chatbot would not have been a serious plan. That is the class of problem the commercial layer still misses, and it is the class Akerman just hired someone to keep hitting.

Zerner’s Job Is the Walk-Away Option

Founded in 1920, Akerman has more than 700 lawyers and business professionals across the United States. It cannot match Kirkland’s $500 million envelope or Latham’s multi-year GPU program dollar for dollar. It can install a person who has already shipped a production review platform and who can tell a practice group that the right answer this month is Copilot, and next month a model tuned on the firm’s own work product.

The June posting asked for a builder. On Sept. 9 the firm hired a commercial litigator who has been in AI since 2000 and in legal machine learning since 2012. Research still buys. Governance still says no. Development is the new fact, and it sits in IT on purpose.

Frequently Asked Questions

What is an open-weight model, and why do law firms want one?

An open-weight model publishes its learned parameters so a customer can download it, adapt it, and run it on hardware the customer controls, unlike a closed API from OpenAI or Anthropic where prompts and outputs travel to the vendor. Latham is fine-tuning Nvidia’s Nemotron 3 family on its own GPUs for that reason, and Akerman said Zerner already built a review platform on open-source models that never left the firm. The draw is less about ideology than about keeping privileged text off a third-party cluster and being able to change models without rebuilding the whole product.

Who holds Akerman’s other two AI director seats?

Michael Adler joined on Aug. 6, 2026 as director of AI governance and data protection and as a partner in the corporate group in Washington, after leading global data privacy and AI at Highspot and working as corporate counsel at Amazon. Tricia Thomas is director of AI and research services, the seat that inventories what vendors already sell. Adler writes the constraints; Thomas maps the catalog; Zerner builds what the catalog does not contain.

What was Vacker v. ElevenLabs?

Voice actors Karissa Vacker and Mark Boyett, with authors and publishers including Brian Larson, Iron Tower Press, and Vaughn Heppner, sued ElevenLabs, a text-to-speech company, alleging it cloned their voices without consent and used the clones as default voices marketed as “Bella” and “Adam.” Munck Wilson filed the case, with Zerner as chief legal architect and a primary drafter, and Akerman said the matter became the first U.S. generative-AI copyright case to reach a settlement. The file sits at the overlap of copyright, personality rights, and the Digital Millennium Copyright Act claims the complaint raised.

How long did Akerman take to assemble this AI leadership bench?

The public steps ran from Jan. 13, 2026, when the firm launched its Law+AI work with USC Gould, through a May retreat of about 700 people, a June 3 job post that demanded a coder, Adler’s Aug. 6 governance hire, and Zerner’s Sept. 9 start, a span of 239 days from the USC launch to the development seat. Meyers had already built Akerman Intelligence as an internal platform for AI strategy and a CEO Perspective series, and the firm already gave lawyers billable-hour credit for AI training. The three directors are the staffing of a program that had been running in policy and training before it had a builder.