The Table · The model can already do it. Why is the product still hard?
Model capability stopped being the interesting constraint. Two teams with access to the same weights ship products that are nothing alike, and the difference is not research. It is interface, defaults, latency, failure behaviour and knowing which twenty percent of the output a user will actually look at.
This table sits on the distance between a capable model and a product people keep using.
On the table
What is left to build when capability is a commodity
How research work survives contact with a product used by millions of non-technical people
Where evaluation quietly replaces the product spec
Our guest: Dr. Eli Brosh — VP AI at Papaya Global, and a StageOne Nexus advisor.
Eli leads AI at Papaya Global, where the product is global payroll and payments and a confident wrong answer is somebody's salary. Before that he was Head of AI Research at Wix, leading the LLM, computer vision and multimodal work behind creative tools used by more than 250 million people, and earlier CTO of Healthy.io and VP Research at Nexar. He has more than thirty publications and patents and a PhD from Columbia, and a decade of moving research into production at consumer scale, which is exactly the distance this question is about.
Eight seats around one table in Tel Aviv: six people building in applied AI product, our guest, and one person from StageOne. 13:00 to 15:00, two hours, one question. An open conversation with someone who has already done it, not a talk: no slides, no presentation, no pitching. Lunch is on StageOne.
Only six seats. Every seat changes the conversation, so if your plans change, please cancel your registration on this page or let us know as early as you can, and we will give the seat to someone else.
Seats are reviewed one by one. We send the exact address in the confirmation email.