← Collected sources
BUILDERS · EDITED DIGEST

Builders’ Picks | 2026-09-20

2026-09-20 · Historical edition

X / Twitter

AI tutorial and interview creator Peter Yang

Peter Yang shared a personal recommendation workflow: he uses a simple Markdown file together with a `/tastemaker` skill to record his ratings for movies and TV shows he has watched and books he has read. Each week, the tool looks for new releases and works he has not yet seen or read, then generates recommendations based on his existing ratings. He said the recommendations matched his personal taste extremely well and asked whether he should open-source the tool. He also previewed a personal finance interview recorded with his friend, the product lead for ChatGPT Finances at OpenAI, centered on six prompts people could try in practice. The uses disclosed so far include proactively finding tax-saving opportunities, maximizing credit card rewards, and generating interactive projections toward financial independence. Both posts focus on having AI use specific personal information to carry out recommendation or planning tasks, but the finance post is currently a preview and does not yet provide the full prompts.

Vercel CEO Guillermo Rauch

Guillermo Rauch believes the most exciting change in the recent AI wave is people’s enthusiasm for building products themselves. He summed up the shift as people wanting to create and ship things, rather than merely consume content. In his view, even amid short-term controversies, the stronger desire to build remains a positive development for the world. His focus is on how the AI wave is changing participation, emphasizing that more people are willing to create things themselves. The post expresses his positive assessment of this enthusiasm for building; it does not discuss the specific capabilities of any tool or provide data to measure the trend.

Box CEO Aaron Levie

Aaron Levie believes personal agents are the fullest expression of the idea of “building what agents want.” Using products such as Muse as examples, he noted that users expect to delegate a task and have the agent complete it from beginning to end, which requires the agent to work smoothly with users’ existing tools or use its own. For tool providers, the capabilities they need to support include invoking MCP or CLIs, browsing websites, and completing transactions. He therefore expects that the attention products need to compete for will increasingly come from agents’ choices. Food ordering, e-commerce transactions, flight bookings, local services, and interactions with data and information are all use cases he explicitly listed. In these scenarios, he believes the tools that are easiest for agents to use and best help them complete tasks will see more usage. He views this shift as the largest opportunity and reshaping of the consumer technology landscape since the App Store.

FPV Ventures Partner Nikunj Kothari

Nikunj Kothari said he spends around 2 hours each day coding, developing side projects, and trying new code repositories, and receives direct messages almost daily questioning why an investor should do hands-on work. His answer is that AI is changing so quickly that without continuous practice, it is hard to understand what will happen next. He divided what AI investors need to understand into three layers: reading technical papers to understand model capabilities and how they are improving; studying infrastructure that supports longer-running tasks; and understanding the application-layer runtime frameworks that enable models to participate in building autonomously operated companies. He particularly emphasized that reading model research papers still offers a substantial informational advantage. Talking with researchers and developers, and assessing founders, are also part of how he learns and invests. He chooses to keep building because he finds it difficult to learn these things solely through secondhand knowledge, and because he enjoys the process. He recommends that anyone seeking to understand AI set aside time for hands-on practice every day, and believes that after trying it for 10 consecutive days, they will likely become absorbed in it too.

SPC General Partner and Bevel Health co-founder Aditya Agarwal

Aditya Agarwal focused on security risks from agent-related intrusions. He mentioned that he had recently seen successive announcements from Gemini (Google), OpenAI, and Anthropic about more agent-related intrusions. These reports led him to a very strong conclusion: some major companies have already been infiltrated by nation-state attackers and remain unaware. He used “100%” to express his certainty, but the post did not identify specific victim companies, describe incidents, or provide direct evidence. This post should therefore be understood as his personal assessment of potential security risks, and cannot establish that any particular company has been breached.