Happy to share my latest post to the GitHub Blog, with “Peli” de Halleux.
Main post: Automate repository tasks with GitHub Agentic Workflows
I’ll add some more personal thoughts – please read the above post first! :)
Over the last year we have been working to bring coding agents to repository automation. This opens a staggering new world of possibilities. I firmly believe this adds “Continuous AI” (CAI) as a third leg to augment “Continuous Integration” and “Continuous Deployment” (CI/CD).
Augment, not replace. Coding agents bring new, magical powers to developers. They are transformative, and increase productivity when well-used. This also applies to repository automation with coding agents. New powers, new opportunities.
All through this work I have advocated that we want developers, teams and communities to be _empowered_ to shape the use of AI-powered repository automation according to _your_ needs, _your_ goals, _your_ responsibilities. That might mean very different things to different people and teams – from all-in to no-use, from code improvement to feature development, from business goals to community goals. That’s for you to shape and decide together.
The automated use of coding agents adds responsibility. This is an offering to contribute to this discussion. Continued work on guardrails and anchoring is essential. This will be a long-running story. The “safe outputs” feature of GitHub Agentic Workflows feels like a step in the right direction. We need more of this.
One of the joys of this work has been to realise that repository automation – when under the control of developers – can be used to make software *better*.
And there is so much “better” to be done. My experience from OSS and enterprise software is that there is a vast amount of _potential_ software engineering to be done. The amount of _potential_ work is almost unlimited, and a lot of it simply isn’t getting done – because time, pressure, cost, a lack of focus. The technical debt is real. The 2,000 production systems in a typical investment bank are real. The need for OSS maintainers to have better tools to get on top of 5000 issues is real. The lack of adoption of strong typing or immutability or better programming languages or latest versions of tooling is real.
Repository automation can also help developers, communities and teams achieve “betterness” that is aligned with core computer science. For example, automated repository agents can improve strong typing day by day. Or make code better by adopting tasteful, helpful use of immutability and functional programming. Or chase performance improvement goals that you define.
This must be under the guidance of talented developers. Your next career move may be to bring previously unreachable levels of quality to companies through guided repository automation. My advice to students at Kings College London will be: become a leader in improving software with AI, make it part of your CV, and learn to do it well.
@dsyme on software 