Adopting the software factory model: crawl, walk, run
The software factory approach (a closed agentic loop that runs in the cloud) is growing in popularity, but it can be daunting to adopt.
Articles
RSS feedAnnouncements, write-ups and release notes from across the AI-devtools ecosystem — linked to the tools and models they're about.
About
September 202652
x.com
Context Infrastructure: Architectural Lessons From the Data Lakehouse
A new pattern is showing up in context infrastructure: store as much of the underlying history as possible, then build the useful representations of it later.
The memory trifecta for personal agents
A useful personal agent should remember your conversations, build lasting knowledge, and learn how you get work done. Take an agent running in Slack as an example.
Software Factories: Building Around a Better Model of Work
Software factories are getting very good at taking on more software work.
Stop thinking about embeddings
@turbopuffer is a search database built directly on object storage that handles both semantic and full-text search.
Ars Umbris: a malleable agent-native IDE for typed knowledge
A knowledge base should work like a codebase: typed files, relationships and conventions an agent understands, checked by a type engine.
Heinrich@arscontextaArs Umbris
How to Build an AI Software Factory: Agents That Open, Review, and Merge PRs
An AI software factory runs fleets of autonomous coding agents that open, review, and merge PRs. How Stripe, Spotify, Shopify, Uber, and Ramp built theirs, broken into five stages you can implement.
Hiba Fathima
Why I open-sourced my software factory
You are not building a software factory. You are building the way your entire company works together.
Building data curation classifiers with agents, SetFit and Jobs
Building a small document classifier with an agent, then using Hugging Face Jobs to label 1% of the English FinePDFs-Edu dataset.
Daniel van Strien
Extending Raschka's GPT-2: an MoE trained from scratch on an RTX 3090
How mixture-of-experts LLMs work, both in theory and in real working code based on 'Build a Large Language Model (from Scratch)'.
Giles Thomas
If coding is solved, what now?: Measuring the sloppiness of code
Exploring how to measure code sloppiness, why correct code can still erode a codebase, and why human intuition and taste still matter.
Sebastian
Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient.
Introducing the smallest model in our new architecture family, with native visual understanding. Designed for greater capability, faster inference, and higher throughput.
Introducing SWE-2: Pushing the Pareto Frontier
Today we’re introducing SWE-2, our most advanced coding model yet. SWE-2 delivers highly competitive agentic coding performance across multiple effort…
The Cognition TeamDevinDevin Desktop
Factoring RSA-260
How Devin and a Cognition researcher built the world’s highest-performance GPU lattice siever, to make factoring numbers 10x cheaper than the previous…
Eric LuDevin
The Complete Guide to pstack Pt. 2
In my last post, I showed you why verification is the foundation of everything I do with agents, and how to get started creating your own verification skills.
AlphaGenome Atlas: molecular predictions for 9 billion human DNA variants
AlphaGenome Atlas: Molecular predictions for 9 Billion human DNA variants — Google DeepMind
Google DeepMind
Augmented Lagrangian Predictive Coding: training 1000-layer networks without backpropagation
A local alternative to backpropagation. PC-ALM trains residual MLPs up to 1000 layers, nearly matching backprop's performance despite using only layer-local dynamics. PC-ALM equips each layer with a feedback control dynamical system that distributes and propagates supervision credit throughout a network.
Jeffrey Seely, Julian Gould
Do it all with Devin: Announcing our Series E
Cognition has raised over $2B at a $48B valuation, led by Andreessen Horowitz and Accel, to build the future of software engineering.
The Cognition TeamDevin
Introducing CUDA Rust: Two Tracks for Writing GPU Kernels
In September 2026, NVIDIA announced it is leaning into native GPU programming in Rust. CUDA C++ and CUDA Python are mature, enterprise-grade toolchains, and NVIDIA will be growing and maturing CUDA…
Sri Koundinyan, Melih Elibol and Jonathan Bentzcuda-oxidecuTile Rust
Introducing OUI-1: world's first model for Generative UI
Full-stack, renderer-agnostic Generative UI with a streaming-first language, official React support, community integrations, and up to 67% fewer tokens than JSON.
OpenUI
Making sovereign, open-weight AI the technology frontier
Mistral today announced that it has raised €3 billion in a Series D funding round at a post-money valuation of more than €21 billion.
Mistral AI
Organizing Context in a Multi-Agent Harness
Learn how context modes in deepagents help subagents fork a supervisor's context or start isolated — for faster, cheaper, more focused multi-agent work.
Thushanth BengreDeep Agents
Astra for Coding: Why Are We Doing This Again?
Some thoughts on Astra and newfangled long-horizon models.
Armin RonacherGPT-6 Astra
Astra vs. The Boys: A Tale of 200 PRs
One Queue bug, an unreasonable number of agents, and 207 pull requests in three days. A story about AI audit, human review, and a GitHub runner quota that gave up.
Michael ArnaldiGPT-6 Astra
Open-sourcing Eraser Diagrams
Eraser Diagrams is an AI-native, coordinate-aware diagram format, open-sourced under the MIT license. This post is about why it needs to exist.
Shin KimEraser Diagrams
An Alien Mind
OpenAI chief scientist Jakub Pachocki argues that rapidly advancing machine intelligence requires stronger alignment and monitoring, scalable defensive systems, safety-gated scaling, and international coordination to keep future AI development under human control.
Jakub Pachocki
Building a Memory-Driven Agent with NVIDIA NemoClaw
Enterprise work spans messages, decisions, projects, and obligations that change over time. An AI agent that starts without this context must reconstruct it before contributing. To provide agents with…
Tanya LenzNVIDIA NemoClaw
Designing an Agent in Slack
Two years of building Windy in Slack: what makes a good Slack agent UX, the details that make it hard, and how Windy, Linear, Mintlify, and Vercel compare.
Max ShawMintlify
I’m running an Opus-level coding agent locally at nearly 2x Claude Opus speed, for free.
Nov 2025 was a tipping point for AI coding with commercial frontier models. Aug 2026 is the beginning of the same thing for local AI. Pretty darned exciting.
Rethinking skills and prompts for GPT-6 Astra
Coding agents have come a long way, and best practices are changing fast. What used to require a lot of handholding and scaffolding no longer does.