Picture it. Moscone West, mid-October, the room is cold because conference rooms are always cold. You have your slides loaded, your demo running on a local build because you do not trust the Wi-Fi, and five investors sitting a few feet away with laptops open. You get your pitch out. Then the questions start, and one of them asks what exactly your agent does when the API call fails. Your co-founder looks at you. You look at the floor.
That is Startup Battlefield 200, and TechCrunch has now named the next group of judges who will be running that exercise at Disrupt 2026, October 13-15 in San Francisco: Nell Daly, Michael Palank, Aditi Maliwal, Chrystal Huang, and Grace Ge.
Why a judge list matters more than it looks
I review AI tools for a living, which mostly means I spend my week watching demos that do not survive contact with a second user. So I have a bias here, and I will state it plainly: the most useful thing about a competition like this is not the prize. It is that a handful of people with money on the line ask questions in public that most founders never have to answer in private.
Product pages do not get interrogated. Launch videos do not get interrupted. A live judging panel does both. When someone who has looked at a few hundred decks asks how your retrieval layer behaves on messy internal documents, there is no editing suite to fix the answer afterward.
That is the part worth paying attention to as a reader rather than a competitor. If you follow AI tooling closely, watching which questions these five ask tells you what investors currently consider unproven. Last cycle it was “is this a wrapper.” Judging panels are a decent leading indicator of where skepticism has moved to next.
What the format actually rewards
Pitch competitions have a known failure mode. They reward performance. A founder who is good on stage can outrun a founder with a better product, and anyone who has watched a few of these knows the feeling of clapping for a team you would not buy from.
But the structure works against that more than people assume, because a panel of five is harder to charm than a panel of one. Different funds, different theses, different pattern libraries of things that went wrong. One judge might be fine with your unit economics while another has already seen three companies die on that exact cost curve. The overlap between five sets of doubts is usually where the real problem lives.
Founders who want to hold up under that should go in with unglamorous answers ready:
- What breaks first when usage multiplies, and what it costs to fix
- Which parts of the product are actually yours versus rented from a model provider
- What a customer does on day 30, not day one
- The specific thing a competitor could copy in a weekend, and why that does not end you
- Where the system fails, stated by you before someone else finds it
That last one is underrated. In my experience the teams that name their own limitations early are the ones whose tools hold up when I test them. Confidence about failure modes is a signal. Confidence about everything is noise.
My honest read for the AI crowd
There is a version of this event that is pure spectacle, and there is a version that is genuinely useful, and which one you get depends on how you show up. If you are competing, treat the panel as free adversarial testing from people whose job is to find the hole in your story. That is expensive feedback normally. Here it is part of the deal.
If you are attending as a builder or a buyer, the demos are the least interesting part of the room. The follow-up questions are where you learn something. You will hear the same three or four concerns surface across unrelated companies, and that pattern is a better map of the AI tooling market than any trend report.
Practical note, since it has a deadline attached: ticket rates go up on September 25 at 11:59 p.m. PT, and registering before that saves up to $200. Standard conference urgency, but the math is the math.
Five investors, a cold room in San Francisco, and a stage where your product either works or it does not. That is a better review environment than most AI tools ever face. I am curious which ones come out the other side still standing.
🕒 Published: