Don Syme, GitHub Next, Visiting Professor, King's College London
In my previous post I introduced Repo Assist: Crunching the Technical Debt with GitHub Agentic Workflows. In my first version of Repo Assist, tasks were selected by round-robin: the agent tracked which tasks it had run most recently and rotated through them. This worked, but it had a problem - the agent was spending equal … Continue reading Adding Weighted Task Selection to a GitHub Agentic Workflow
I have spent the weekend using GitHub Agentic Workflows to make Repo Assist, a generic, all-purpose automated repository assistant for software maintainers. For some repos, some of the time. I crafted RepoAssist for my own purposes, for helping with repos with long term technical debt. These are repositories where: The project is valuable ✨ You … Continue reading 🌈Repo Assist: Crunching the Technical Debt with GitHub Agentic Workflows
People using "coding agents in a loop" have discovered the joy of implementing programming language compilers using this technique. It's a lot of fun it's impressive how much can be achieved. I dug out my notes from an experiment (really a "spike") we did in July 2025. Aim: Implement a programming language for teaching Approach: … Continue reading Spike from July 2025: Creating a Programming Language using Coding Agents on GitHub
In previous posts I've written about Natural Language Programming, and Intent/Realization Toolchains as two grounding concepts in AI productivity tools, particularly AI for Code. In this post I want to explore another angle that comes up in features in AI productivity tools, particularly those that generate documentation, synthesize specifications, discover build rules or other "extractive" … Continue reading Generative AI and Changing Inputs
(This blog post is written in personal tone, but relates to our work at GitHub Next and may be moved to https://githubnext.com in future. A huge thank you to Peli de Halleux, Joe Zhou, Eddie Aftandilian, Russell Horton, Idan Gazit and many others at GitHub Next, and the GitHub platform leadership of Mario Rodriguez. I'm … Continue reading Towards Semi-automatic Agentic Performance Engineering
In previous posts I've written about Natural Language Programming, Dijkstra's Ghost - the End of The Symbolic Supremacy and Ephemeral Editable Specifications (aka Extract, Edit, Apply). These touched on the topics of Natural Language Programming and the role of Specifications in AI-native programming. Today I'd like to step back and address an underlying question: what … Continue reading What Kind of Programming is Natural Language Programming?
Ever since we started working on "task-oriented programming" (aka vibe coding) in 2023, our group at GitHub Next have been throwing around ideas related to "continuous" tasks in software repositories: Continuous Code Cleanup, or Continuous Documentation and so on. This finally bubbled up as the Continuous AI project, locating it within the tradition of Continuous … Continue reading On Continuous AI for Test Improvement
I'm excited to share our latest research demonstrator from GitHub Next - "GitHub Agentic Workflows - Natural Language Programming for GitHub Actions" .https://githubnext.com/projects/agentic-workflows/Agentic Workflows focuses on expressing repository‑level behaviors in natural language and running them on GitHub. Agentic Workflows is not a product and not even a technical preview; it's a vehicle for exploring the agentic design space, … Continue reading GitHub Agentic Workflows
Sharing a write-up of one of our investigations at GitHub Next: Extract, Edit, Apply. Spec-oriented programming is usually seen as "Spec-first", with a compilation step to turn specs into code: Specs are permanent, and Code is ephemeral. This has many obvious problems, including: The instability of LLM code-generation under otherwise small or unimportant changes to … Continue reading Extract, Edit, Apply – a design pattern for AI
GPT-4 and other LLMs (Large Language Models) are driving a tidal wave of innovation in applied AI. However used without augmentation they have very limited calculational capabilities and make mistakes calculating with numbers. In this project, we describe a simple, general technique to address this, apply it to some widely reported real-world failures of GPT-4-based … Continue reading Augmenting GPT-4 with Calculational Code
@dsyme on software 