Among the surveyed AI engineers, cost was the top consideration when choosing a model.
2
The most popular answer for an AI buzzword people are tired of hearing was AGI, followed by RAG and prompt engineering among the other answers.
3
Attention Is All You Need was the most popular answer for the most influential AI paper in history.
Summary
Barr Yaron hosts a Family Feud-style competition between engineers from Anthropic, Reflection AI, Google DeepMind, Augment Code, Google, and Thinking Machines. The teams guess answers collected from 100 AI engineers, with points based on popularity. The survey questions cover influential researchers, model selection, overused AI terms, tools engineers like, jobs exposed to AI disruption, influential papers, late-night engineering problems, and industries that could benefit from AI. The answers offer a quick snapshot of what this audience thinks. Cost ranks first when choosing a model. AGI is the most disliked buzzword. Cursor is the AI tool engineers love most, while Attention Is All You Need is the top paper. The event also previews a larger State of AI Engineering survey about tools and workflows, which Barr says will be presented at the AI Engineer World Fair.
The game uses popularity among 100 AI engineers as its scoring system
Barr explains that the teams are not trying to find objectively correct answers. The organizers surveyed 100 AI engineers on a series of questions, then assigned the most points to the answers that appeared most often. Audience members can guess along with the teams. The first question asks for the most influential AI researcher. The board includes Yann LeCun, Andrew Ng, Fei-Fei Li, and Yoshua Bengio, while guesses such as Ilya Sutskever, Noam Shazeer, Sam Altman, and Jeff Dean do not make the top eight.
The model-selection round produces a result Barr says surprised her: cost is the number one consideration. The other answers include latency, evaluation or benchmark scores, and accuracy or performance. The teams also suggest safety, where the model is served, and open source versus closed source, but those do not appear in the scored answers. The result gives a practical picture of model buying decisions. Price comes before several technical properties that the contestants initially expect to matter most.
AGI is the buzzword engineers are most tired of hearing
The teams are asked to name an AI buzzword everyone is tired of hearing. AGI is the number one answer. Other guesses include agents, multimodality, copilots, DeepSeek, and safety. The additional answers shown on the board are RAG and prompt engineering. The contestants treat some terms as changing quickly, with DeepSeek described as perhaps being tiresome only for the current week. The round gives a blunt view of which parts of AI vocabulary have become repetitive for working engineers.
The winning team takes the game with a mixture of survey answers and hot takes
Roo's Basilisk wins the main game after the buzzword round. The team includes Paige and Mahir, and Barr credits their performance to attention and the mixture of experts. The final round asks the winners to reach 200 points through five rapid questions. Mahir earns 140 points, helped by Cursor, software engineering, Attention Is All You Need, and hardware failure, which Barr accepts as an information problem. John then answers with Claude, artists, and a CUDA issue, but the team falls short of 200.
Cursor is the most popular AI tool among the surveyed engineers
In the fast-money round, the question about an AI tool engineers love produces Cursor as the number one answer. The second representative answers with Claude, and the board also includes model APIs. Barr later displays the top answers again, with Cursor at the top. The result is one of the clearest signals from the survey because both the speed round and the final board make the tool question concrete. The contestants' other answers show that the survey covers tools beyond model providers.
Attention Is All You Need leads the paper question by a wide margin
Attention Is All You Need is the most popular answer to the question about the most influential AI paper in history. Mahir gives the answer first, and John repeats it after hearing the question. Barr calls it by far the number one answer when revealing the board. The fast-money board also lists it as the top paper, alongside questions about AI-disrupted jobs and engineering problems at 2 a.m. The contestants' answers include software engineering, artists, content creation or writing, and hardware failure.
Paige expects deployed models to move toward smaller on-device systems
During introductions, Paige gives a forecast about deployed AI models. She expects that, a year and a half later, most deployed models will be on device. She also expects more smaller models, orchestrated together and specialized for particular tasks, instead of relying only on larger models and sending data elsewhere. Her view is part of the event's opening hot takes rather than one of the scored survey answers. Other panelists predict biological materials for future model architectures, AI systems acting as therapists, and conversations involving several bots.
The organizers plan a broader survey about AI engineering work
Barr closes by inviting viewers to take part in a larger State of AI Engineering survey. She says it will cover the tools engineers use and their workflows, with results presented at the June AI Engineer World Fair. She describes the survey as a way for the industry to be more transparent. The questions used in Frontier Feud are presented as teasers for that work. The closing invitation points viewers to a QR code or link rather than offering more analysis of the game.
"The goal is for the two of them together to get to 200 points."16:50
Who should watch
You want a quick view of what 100 AI engineers ranked highly when choosing models, tools, papers, and AI use cases.
You are comparing the assumptions engineers from frontier labs bring to questions about model deployment and AI coding tools.
You want practical survey signals without watching a full technical presentation.
