Agentic Workflows with Don Syme
1Digest
This podcast explores the transformative impact of agentic workflows and AI on software development. Don Syme, designer of F#, discusses how modern coding agents are empowering developers, akin to wizards. He reflects on the evolution of software craftsmanship, comparing new tools to a chainsaw replacing an axe, emphasizing the enduring pursuit of quality software. The conversation highlights practical applications, such as a trivial app example illustrating the full development lifecycle, and how AI agents can proactively improve engineering practices like trusted publishing. Concepts like "ambiguity loops" are introduced as flexible decision-making mechanisms in AI, distinct from traditional loops. "Continuous AI" is presented as a new pillar in the SDLC, complementing CI/CD with automated documentation, triage, and code simplification. The episode addresses skepticism about AI by emphasizing integrity and developer empowerment through tools like GitHub Agentic Workflows. These workflows, powered by AI agents processing markdown prompts in sandboxed environments, are contrasted with deterministic YAML workflows. A success story, "Repo Assist," demonstrates how agentic workflows can manage technical debt, label issues, fix bugs, and prepare releases for open-source projects, re-energizing maintainers. The interaction model involves developers checking progress made by agents. Limitations and best practices, including the need for human oversight, benchmarking, testing, and robust "harnesses" and guardrails to prevent AI-generated "slop," are discussed. Finally, the importance of educating future engineers in AI-assisted development is stressed, concluding with a call to explore repository automation.
Outlines
Introduction to Agentic Workflows and the Evolution of Software Development
Scott Hanselman introduces Don Syme, who discusses the current state of computer science and the evolution of software development, comparing new AI-powered tools to advancements like a chainsaw replacing an axe. They explore how modern coding agents empower new engineers and the ongoing pursuit of reliable software.
Practical Applications and Improving Engineering Practices with AI
An example of a simple Windows app illustrates the software development lifecycle, including GitHub workflows, certificates, packaging, and updates. The discussion highlights how AI agents can proactively suggest improvements, such as using trusted publishing and empowering repository maintainers.
Advanced Concepts: Ambiguity Loops and Continuous AI in the SDLC
The concept of "ambiguity loops" is introduced as a flexible decision-making process for AI, distinct from deterministic loops. "Continuous AI" is presented as a third pillar alongside CI/CD, encompassing continuous documentation, triage, and code simplification.
GitHub Agentic Workflows: Explanation, Benefits, and Success Stories
Addressing AI skepticism, the focus shifts to GitHub Agentic Workflows, which use markdown prompts processed by AI agents in a sandboxed environment. These differ from deterministic YAML workflows. "Repo Assist" is highlighted as a success story, helping manage technical debt, label issues, fix bugs, and prepare releases for open-source projects, re-energizing maintainers.
Limitations, Best Practices, and Future of AI in Software Engineering
The conversation acknowledges limitations and the necessity of human oversight, emphasizing the importance of benchmarking, testing, and implementing "harnesses" and guardrails to ensure AI reliability. The need for educational institutions to offer courses on AI-assisted software development is stressed, concluding with encouragement to explore repository automation.
Keywords
Agentic Workflows
A new paradigm in software development where AI agents autonomously perform tasks within a defined context, guided by human-defined goals and constraints, enhancing efficiency and capabilities in the SDLC.
Continuous AI
An extension of CI/CD principles, integrating AI capabilities throughout the software development lifecycle for tasks like automated documentation, code simplification, and continuous improvement.
Ambiguity Loops
A flexible decision-making process in AI systems that allows for dynamic problem-solving and adaptation, contrasting with deterministic programmatic loops, crucial for robust AI-driven software.
Repo Assist
A specific agentic workflow tool designed to assist open-source maintainers by automating tasks like issue management, bug fixing, and release preparation, reducing technical debt and improving repository health.
Software Development Lifecycle (SDLC)
The entire process of developing software, from planning and creation to testing and deployment. Agentic workflows and Continuous AI are presented as advancements to optimize and enhance the SDLC.
GitHub Agentic Workflows
GitHub's implementation of agentic workflows, utilizing markdown prompts processed by AI coding agents in a sandboxed environment to automate and improve software development tasks.
AI in Software Engineering
The application of artificial intelligence techniques and tools, such as agentic workflows and continuous AI, to enhance the efficiency, quality, and automation of the software development process.
Q&A
What are agentic workflows and how do they differ from traditional YAML workflows in GitHub?
Agentic workflows utilize markdown prompts processed by AI coding agents within a sandboxed repository environment. They are intent-based and focus on flexible decision-making, unlike deterministic YAML workflows which are more script-like.
How does "Continuous AI" fit into the software development lifecycle?
Continuous AI adds a new dimension to CI/CD, enabling ongoing automation for tasks like documentation updates, code simplification, security analysis, and testing, thereby continuously improving the software.
What is "Repo Assist" and how can it help open-source maintainers?
Repo Assist is an agentic workflow that automates repository maintenance tasks. It helps manage technical debt, label issues, propose code improvements, and prepare releases, significantly easing the burden on maintainers.
What are "ambiguity loops" and why are they important in AI software development?
Ambiguity loops provide AI with flexibility to make decisions and solve problems dynamically. They are essential for creating more robust and adaptable software, but require careful implementation with checks and balances.
How can developers ensure AI-generated code or suggestions are reliable and safe?
Reliability and safety are ensured through strong guardrails, human oversight, and rigorous testing. Agentic workflows operate in read-only modes initially, with results checked by humans before implementation.
Show Notes
In this episode, Scott talks with Don Syme about the emerging world of agentic developer workflows and what it means when coding tools move from autocomplete helpers to collaborators. They explore how modern tools like GitHub Copilot and GitHub Agentic Workflows are evolving into systems that can plan, execute, and iterate on tasks across a codebase, and what that means for software design, type systems, and developer responsibility.
Table of contents
United States
