From C Student to Valedictorian: What AI’s Law School Journey Teaches Us About Legal Technology’s Future

Two and a half years ago, an AI walked into a law school classroom and stumbled out with grades that would have gotten me a stern talking-to from parents and my professors. Fast forward to Spring 2025, and the latest AI is earning A+ grades. Certainly better than most of mine.

As someone who’s sat in those same exam rooms sweating over Constitutional Law hypotheticals, and who’s now spent nearly two decades building AI systems for legal use cases, this remarkable journey hits differently for me. It’s not just academic progress—it’s validation of something I’ve believed since we first started using natural language processing back in 2004: AI is going to fundamentally transform the practice of law.

My Own Journey: From Law Student to AI Builder

When I was grinding through law school exams, I never imagined I’d one day be building systems that could outperform me and every other student on those same tests. But here we are. The progression documented in these studies, from GPT-3’s C- grades to o3’s A+ performances, mirrors the rapid evolution of basic GPT models to today’s uber intelligent reasoning models.

In Fall 2022, researchers had ChatGPT take law school finals, where it earned grades no higher than a B. By Spring 2025, OpenAI’s o3 model didn’t just pass—it earned three A+ grades, outscoring top students in Constitutional Law, Legal Profession, and Property.

Watching an AI go from struggling student to valedictorian candidate in less time than it takes to complete law school? Remarkable.

The Breakthrough That Changes Everything

Here’s what really gets me excited—and what directly influences how we’re building The Contract Network. It’s not just that o3 got better grades. It’s how it thinks.

Unlike earlier models that generate responses instantly, o3 is a “reasoning model” that uses internal deliberation to test multiple approaches before settling on an answer. Sound familiar? That’s exactly how the legal community is taught to solve problems. The IRAC method (Issue, Rule, Analysis, Conclusion), the careful consideration of multiple theories, the iterative refinement of arguments–o3 is doing all of this internally.

This isn’t just incremental improvement. This is AI learning to think like a legal professional.

For those of us building legal technology, this architectural shift is game-changing. These reasoning models aren’t just pattern-matching machines anymore. They’re engaging in something that looks remarkably like legal reasoning. And they’re on track to do so in a fashion that exceeds the performance of any other law school student.

We’ve All Become Prompt Engineers

One finding from these studies provided powerful validation of something we’ve known since the day we incorporated AI into our product. When researchers tweaked a single prompt to ask GPT-4 to “spell out your reasoning and calculations… in painstaking detail,” its Income Tax exam performance jumped by nearly 50%.

This perfectly illustrates why some of our highest-performing prompts for intricate tasks average over 10 pages in length. It’s something we’ve been teaching in workshops at my law school on best practices in prompting—that “product level” prompting is fundamentally different from ad hoc instructions. The precision and detail matter enormously. Because even the most sophisticated AI is only as good as the instructions we give it.

More Isn’t Always Better (A Lesson We Built Around from Day One)

Here’s a finding that validated something we’ve architected around since day one: giving AI access to 70,000 words of class notes actually hurt its Torts exam performance.

Having worked with AI over nearly two decades, we understood early that AI systems, like legal professionals, perform best with focused, relevant information rather than overwhelming data dumps. This deep expertise in legal AI shaped how we built The Contract Network from the ground up.

This is why our platform emphasizes focused, relevant context over comprehensive data dumps. When we help parties negotiate agreements, we don’t throw every possible precedent at the problem. We carefully curate the most relevant information. Quality over quantity—a lesson that applies equally to human professionals and AI systems.

The Promise and the Peril

Despite o3’s stellar grades, the studies revealed weaknesses that mirror challenges I see every day. Earlier models missed crucial issues, provided shallow analysis, and failed to follow standard legal reasoning. In one particularly concerning case, GPT-4’s criminal law analysis was so flawed it “could have resulted in the client going to jail.” These problems have become less prevalent as reasoning models have emerged, but some of the flaws persist. For example, this spring the AI missed a landmark case that was critical to an exam. The reason? o3’s training data had a knowledge cut-off four weeks before the decision that fundamentally changed the subject was released.

What This Means for Our Future Together

These studies aren’t just academic curiosities. They’re a roadmap for where legal technology is headed—and where The Contract Network is going.

When I see AI evolving from C- to A+ performance, I see validation of our core belief: the future of contracting lies in the synergy between human judgment and AI capabilities. We’re not building systems to replace negotiation professionals, but we are building systems that make them exponentially more effective.

Imagine contract negotiations where AI handles the routine analysis and markups, while negotiators focus on strategy and relationship-building. Picture a world where every party—regardless of size or resources—has access to sophisticated legal analysis and to speed. This is the future we’re building, one agreement at a time.

A Personal Invitation

If an AI can go from struggling student to honor roll in 2.5 years, what will the next 2.5 years bring? Having spent nearly two decades at the intersection of law and AI, I can tell you this: the pace of change is accelerating at an unprecedented pace, and the opportunities are extraordinary.

The law school classroom has become an unexpected laboratory for understanding AI’s capabilities. I find these developments both humbling and inspiring. These AI breakthroughs bring us closer to democratizing access to high-quality legal support in ways previously unimaginable.

Join us on this journey. Together, we can ensure that as AI gets smarter, our legal system gets not just more efficient, but more just.

Jim Wagner is CEO and Co-founder of The Contract Network. He is a serial entrepreneur in the legal technology community and a recognized leader in artificial intelligence for contracting, having analyzed tens of millions of contracts with AI systems over the past 20 years.

Studies of Interest:

¹ Choi, J. H., Hickman, K. E., Monahan, A., & Schwarcz, D. B. (2023). ChatGPT Goes to Law School. Journal of Legal Education, 71, 387. https://ssrn.com/abstract=4335905

² Blair-Stanek, A., Carstens, A., Goldberg, D. S., Graber, M., Gray, D. C., & Stearns, M. L. (2023). GPT-4’s Law School Grades: Con Law C, Crim C-, Law & Econ C, Partnership Tax B, Property B-, Tax B. https://ssrn.com/abstract=4443471

³ Blair-Stanek, A., Campbell, P., Gifford, D. G., Krishnamurthi, G., Percival, R. V., Sovern, J., Sweeney, M., Tobin, D. B., & Vertinsky, L. (2024). GPT Gets Its First Law School B-pluses. https://ssrn.com/abstract=4717411

⁴ Blair-Stanek, A., Gifford, D., Graber, M., Krishnamurthi, G., Sovern, J., Tobin, D., & Van Alstine, M. (2025). AI Gets Its First Law School A+s. University of Maryland Carey School of Law. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5274547