Blog
OpenAI Model Disproves Major Conjecture in Discrete Geometry: A New Era of Mathematical Discovery by AI
An OpenAI model has successfully disproven a central conjecture in discrete geometry. We explain this case of a mathematical breakthrough achieved through "reasoning" that goes beyond simple computation.
[Case Study] How Ramp Used Codex to Accelerate Code Reviews
We introduce a case study of the fintech company Ramp, which dramatically streamlined its code review process using OpenAI's Codex. We explain the approach that reduced review tasks from several hours to a few minutes.
How to Ensure Trust in AI-Generated Content: The Importance of Content Provenance
As AI-generated content becomes increasingly indistinguishable from human-created work, ensuring its authenticity through content provenance is vital. Technologies such as invisible watermarks and industry standards like C2PA offer practical solutions, urging developers to embed these measures for a transparent and trustworthy AI ecosystem.
How AI-Native Companies Are Redefining Organization Design: From Efficiency to Capability and Self-Evolution
Y Combinator insights reveal that AI-native companies focus on enhancing individual capabilities rather than just efficiency, transforming organizations into self-improving systems. Learn the key principles, from making companies queryable by AI to building closed-loop systems that evolve automatically.
How to Build a Self-Improving AI Company: Insights from YC's Latest Startup Guide
Y Combinator partners argue that the next generation of startups must shift from using AI for mere productivity gains to unlocking new, superhuman capabilities. The core idea is to transform companies into 'self-evolving AI loops' by making all information machine-readable, eliminating hierarchical 'human routers,' and building closed systems where AI learns from its own actions to improve the business overnight.
Elon Musk's $10 Billion Bet: Why Coding Agents Are the New Frontier in AI
Elon Musk's clash with Anthropic over AI model access revealed a critical insight: coding agents are not just tools but essential data flywheels for training superior AI models. This discovery has sparked a race among AI giants to build their own coding products, reshaping the competitive landscape.
Building an AI-Native Company: Prioritize Capability Over Productivity to Create a Self-Evolving Organization
This article argues for a fundamental shift in thinking when building AI-native companies, moving beyond using AI for productivity gains to leveraging it for exponential capability expansion. It outlines a framework for transforming organizations into self-improving systems where AI handles information processing and coordination, allowing human roles to evolve towards judgment and real-world engagement at the 'edge'.
How to Build a Self-Evolving Company With AI: Lessons From Y Combinator's Latest Lectures
Based on recent lectures by YC partners Tom Blomfield and Diana Hu, this article lays out a radical vision for AI-native companies: organizations should become self-evolving systems where AI reads, queries, and continuously improves the company's knowledge and processes. The key insight is that AI's real promise isn't incremental productivity gains — it's enabling a single person to achieve what entire teams once could, by transforming the company itself into a closed-loop, self-improving machine.
Claude Subagent Explained: How to Keep Your Context Window Clean and Speed Up Complex Tasks
Claude Code's Subagent system lets you offload heavy tasks into independent context windows, keeping your main dialog clean and enabling parallel execution. This guide covers what Subagents are, the three built-in options, how to create custom ones, and practical scenarios where they shine — from test runs to chained workflows.
Frontiers of AI Agent Design: System Architecture and Technical Trends Report for 2026
The design philosophy for AI agents heading into 2026 is shifting from "all-purpose single models" to "configurable, observable, and constrained systems." This report details the separation of planners and executors, trajectory-based evaluation, and the latest architectures to increase real-world robustness.