Monday — September 1, 2025

A third of senior developers say over half their code is AI-generated, researchers propose a background-independent algebra in quantum gravity, and Pitaya is introduced as an AI coding agent orchestrator that runs multiple agents in parallel.

News

Survey: a third of senior developers say over half their code is AI-generated

A survey of 791 developers found that senior developers are more likely to use AI-generated code in production, with 32% saying over half of their shipped code is AI-generated, compared to 13% of junior developers. Despite some drawbacks, such as the need to edit AI-generated code, 59% of senior developers and 49% of junior developers say AI tools help them ship faster overall, and nearly 80% of developers say AI tools make coding more enjoyable.

No clicks, no content: The unsustainable future of AI search

The rise of AI companies like ChatGPT and Google is causing a content drought as they divert traffic away from publishers and businesses, threatening their survival and potentially starving themselves of the content they rely on to train their models. The disruption of the online economy may ultimately lead to a collapse, potentially through an economic bubble bursting, or may require regulation to address the issue of AI companies profiting from scraped content without fairly compensating creators.

Are people's bosses making them use AI tools?

The author of Piccalilli spoke with several developers and designers who shared their negative experiences with being encouraged or forced to use AI tools by their bosses, leading to frustration, fear, and a sense of unease in the industry. These conversations revealed that some companies are pushing AI use to cut corners and increase profits, with some managers even threatening to replace employees who don't adapt to AI tools, ultimately creating a toxic work culture.

ETFs now hold more than $3.1T worth of just top US companies

The top 10 equity holdings by US-listed ETFs have a total market value of $3.1 trillion, with the top holdings including NVIDIA, Microsoft, Apple, Amazon, and Alphabet, among others. The full list of top 100 holdings includes a diverse range of companies across various industries, with market values ranging from $612 billion for NVIDIA to $16.3 billion for Doordash.

Zuckerberg’s AI hires disrupt Meta with swift exits and threats to leave

Here is a 2-sentence summary of the text: Meta CEO Mark Zuckerberg is leading a major reorganization of the company's senior leadership, shifting away from longtime executives and towards a new generation of AI-focused leaders, including recent hires like Shengjia Zhao and Alexandr Wang. The transition has been tumultuous, with some new hires already leaving the company and others struggling to adapt to Meta's bureaucracy and internal competition for resources, as Zuckerberg pushes to accelerate the development of "personal superintelligence" and catch up with rival AI labs.

Research

Navigating the Latent Space Dynamics of Neural Models

Neural networks can be viewed as dynamical systems acting on a lower-dimensional latent space, where autoencoder models define a latent vector field that emerges through standard training procedures. This vector field representation enables analysis of model properties, such as generalization and memorization, extraction of prior knowledge, and identification of out-of-distribution samples, and has been validated on vision foundation models.

A Background Independent Algebra in Quantum Gravity

An algebra of operators is proposed as a background-independent framework in quantum gravity, utilizing the Hartle-Hawking no boundary state as a universal state of maximum entropy. This approach yields sensible results when applied to de Sitter vacua with varying cosmological constants, allowing for the definition of entropy in terms of relative entropy with the universal state.

A Conceptual Introduction to Hamiltonian Monte Carlo [pdf]

Hamiltonian Monte Carlo has achieved significant empirical success, but its underlying mathematics, rooted in differential geometry, have limited its understanding and dissemination to broader communities. This review aims to provide a comprehensive and intuitive account of the method's theoretical foundations, allowing practitioners and statisticians to gain a solid understanding of how Hamiltonian Monte Carlo works, its strengths, and its limitations.

Code

Show HN: Pitaya – Orchestrate AI coding agents like Claude Code

Pitaya is an AI coding agent orchestrator that runs multiple agents in parallel, allowing users to pick the best result, and supports custom Python strategies for multi-stage workflows. It features a clean, adaptive terminal user interface and supports plugins like Claude Code and Codex CLI, with built-in strategies for tasks such as scoring, iterative refinement, and bug finding.

Show HN: 70 days, 800 stars. If AI bugs are not random but math inevitable?

The WFGY project is a semantic reasoning engine designed to solve core AI problems such as hallucination, context drift, and logic failure, with a goal of igniting a new civilization layer built on semantic reasoning. The project includes various modules, such as TXT OS, Blah Blah Blah, and Blur Blur Blur, which provide functionalities like semantic Q&A, image generation, and reasoning games, all running natively as .txt apps without installation or dependencies.

Curated list of 500 AI Agents and their use cases across industries

This repository provides a curated collection of over 500 AI agent use cases across various industries, including healthcare, finance, education, and more, with links to open-source projects for implementation. The use cases range from personalized learning and financial trading bots to smart farming assistants and energy demand forecasting agents, showcasing the diverse applications of AI agents in transforming industries.

Show HN: Claudable – open-source web builder powered by 5 CLI AI coding agents

Claudable is a powerful Next.js-based web app builder that uses AI coding agents like Claude Code and Codex CLI to generate production-ready code and deploy apps instantly to Vercel. It offers features like natural language to code, instant preview, and zero setup, allowing users to build and deploy professional web applications for free with optional integration with Supabase database and GitHub.

New RAG algorithm – Heuristically densifying data into many perspectives

SpinRAG is a Retrieval-Augmented Generation algorithm that acts as a low-level brain for Large Language Model pipelines, providing fast and dynamic context and memory through an evolving knowledge graph. The algorithm uses a unique "spin" concept to model data interactions, with four spin types and production rules that drive the graph's evolution, allowing for efficient and contextually rich querying and retrieval of information.

    A third of senior developers say over half their code is AI-generated, researchers propose a background-independent algebra in quantum gravity, and Pitaya is introduced as an AI coding agent orchestrator that runs multiple agents in parallel.