I Prepped for State Debate Finals Using AI. My Coach Wasn't Happy About What I Learned.

Three weeks before state finals, my coach said something that stuck with me: "You're good at the arguments you've practiced. But you haven't faced an opponent who's read everything."

She was right. Our team practiced against each other, which meant we faced the same arguments on repeat. I could counter the cases my teammates made because I'd seen them a dozen times. But at state, I'd face debaters from across Texas who'd found angles I'd never encountered.

My coach's solution was to read more — pull more cards, find more sources, build a bigger evidence file. That's the standard approach. The more you've read, the less surprised you'll be.

I did something else. I started practicing against AI.

---

The Problem with Human Practice Partners

Here's the limitation of practicing against your teammates: they have the same constraints you do. Same research time, same source access, same biases about which arguments are "good." When you debate someone who prepared the same way you did, you don't learn what you don't know. You just get better at the arguments you've already found.

My state topic was "The United States should substantially increase its security cooperation with NATO." I'd prepped the standard positions. Pro: collective defense, deterrence, burden-sharing. Con: overextension, provoking Russia, free-rider problem.

My teammates ran variations of those same positions because those are the arguments that show up in camp files and debate briefs. We were all pulling from the same playbook.

The AI didn't have a playbook. It had everything.

---

Round 1: The Argument I Should Have Found

I ran my 1AC (first affirmative constructive) against DebateAI. Standard case: increasing security cooperation with NATO deters Russian aggression and prevents war.

The AI's response wasn't what I expected. It didn't attack my evidence or argue that deterrence doesn't work. Instead, it attacked something structural:

"Your case assumes 'security cooperation' means increased military presence. But the resolution says 'substantially increase security cooperation' — that could include intelligence sharing, cyber defense, or diplomatic coordination, none of which require troops or bases. You've chosen the most escalatory interpretation of a resolution that permits softer options. Why? If your goal is deterrence, there are less provocative means. If your goal is something else, you haven't stated it."

I'd never had an opponent question the framing of my interpretation. In practice rounds, we all assumed "security cooperation" meant military buildup because that's what the evidence supported. The AI noticed I was arguing for a specific version of the resolution without justifying why that version was best.

My coach hadn't caught that. My teammates hadn't caught it. The AI found it in three minutes.

---

Round 2: The Counterargument That Made Me Rewrite My Case

The AI then ran a position I'd never encountered: the "restraint" critique.

"Increased U.S. commitment to NATO security enables European free-riding on defense, preventing Europe from developing independent military capacity. This makes the alliance structurally dependent on American power, which is unsustainable given domestic U.S. politics and budget constraints. Your case makes the problem worse: more U.S. commitment now means less European self-sufficiency later, which means a more fragile alliance when U.S. priorities shift. You're solving short-term deterrence by guaranteeing long-term collapse."

I'd heard the "free-rider" argument before, but always framed as a reason to get Europe to pay more. The AI flipped it: maybe free-riding wasn't a bug to be fixed but a symptom of a structural problem that more commitment would deepen.

That reframe changed my entire case. I added an answer to the restraint critique that became one of my strongest blocks. I also prepped to run the restraint case on negative, which nobody in my district was doing.

---

Round 3: The Rhetorical Problem I Didn't Know I Had

A few days later, I ran cross-examination practice against the AI. I played the questioner; it played the affirmative, defending increased drone cooperation with NATO.

AI: "Expanding drone surveillance sharing with NATO allies improves threat detection without forward deployment of troops."

Me: "But doesn't that increase the risk of misidentification errors, like we've seen with U.S. drone strikes in Pakistan and Yemen?"

AI: "You've shifted from 'sharing intelligence' to 'strike operations.' My case is about surveillance cooperation, not strike authority. NATO allies receiving better intelligence doesn't mean NATO allies are conducting U.S.-style drone warfare. You're conflating two different uses of the technology."

I'd thought I had a killer question. The AI showed me I was equivocating between intelligence sharing and kinetic operations — two things that sound similar but have completely different policy implications. If I'd asked that question at state, a sharp opponent would have exposed the slippage and made me look sloppy.

My coach taught me how to structure CX questions. The AI taught me how to avoid questions that sound good but don't actually apply.

---

What the AI Couldn't Do

Let me be clear: AI practice didn't replace anything. It added something.

The AI couldn't tell me when I was speaking too fast. It couldn't read the room, notice when a judge was confused, or adapt to the other team's style. It couldn't simulate the pressure of a real round or the chaos when someone runs a position you've never heard of and you have five minutes to respond.

Those things require human practice. There's no substitute for rounds against real people with a real judge.

What the AI could do was expose gaps in my arguments that human practice partners missed — not because they're worse debaters, but because they're biased in the same ways I am. They find the same sources, build the same cases, and face the same time constraints. The AI had none of those limitations. It just made the best available argument, whether I'd encountered it before or not.

---

State Finals

I made it to elimination rounds. Lost in semifinals to a team from Houston running a kritik I'd never seen — some kind of security studies argument I still don't fully understand. That's debate.

But here's what I noticed: in every round before that, I faced fewer surprises than my opponents faced. When someone ran a standard argument, I had answers they weren't expecting. When I ran the restraint critique on negative, two of my opponents looked like they'd never heard it. They probably hadn't — it wasn't in the camp files.

The AI didn't teach me that case. But it introduced me to the framing, which led me to the literature, which led me to the argument. That pipeline — AI surfaces an angle, you go find the evidence — turned out to be faster than starting from scratch in the card pile.

---

What I'd Tell Other Debaters

If you're prepping for a tournament and you only practice against your teammates, you're training for opponents who think like you. That's not who you'll face in elimination rounds.

Use AI to find the arguments you don't know you're vulnerable to. Set up a debate on DebateAI, run your case, and see what comes back. Some of it will be things you've heard before. Some of it will be garbage. But every few rounds, the AI will find something real — a framing problem, a dropped objection, an angle nobody in your prep group considered.

That's worth an hour of your time.

My coach still thinks I should have spent more time reading cards. She's probably right — you can never have too much evidence. But I also know that the AI found things the card pile didn't. Both matter. The students who figure out how to use both are the ones who'll be hardest to beat.