Contributed by Kris Buytaert
Is trunk-based development still possible in an agent-assisted world?
Pull requests within internal teams have always been one of the best ways to block a team’s flow. Teams that still use internal pull requests for the codebase they own have trust issues to deal with; they need to invest in leveling up the skills of their team members by practicing pair programming or finding other ways to eliminate that lack of trust.
The main place where PRs have value is in open-source projects, or in internal projects when new team members join. In those cases, it makes sense to let them work through a PR-type workflow for a while, until both the team and the new members reach a level of trust that what they are doing is right.
However, at what point can you trust LLMs to be reliable team members? What do you need to put in place to reach that goal? And while you aren’t there yet, you have to resort to treating them as a resource you can’t fully trust.
Unlike a new human joining the team, an LLM will simply DDoS you with PRs… The amount of code they generate in a short time span is massive. We already see teams struggling with this—and that’s just the functional code.
The real challenge is that because of LLM-assisted code, some teams will move faster in what they are currently doing, delivering more artifacts that need to be deployed, monitored, scaled, and decommissioned. Other teams will also increase their pace, but their code quality might decrease, leading to a higher rework rate.
In a true DevOps ecosystem, the team will need to take ownership of more and more artifacts. In a more legacy DevOps approach, the ops team will need to keep up with the pace, complexity, and variety of those artifacts. Once again, the only solution here is a single automated path to production—which requires intensive automation, standardization, and increased quality control. The challenge is that most teams still operating in this model are in denial about the current state of their automation.
What tools do we have out there that would allow us to rebuild that trust? How does the current set of tools fit into that ecosystem?
As the Continuous Delivery Community, this is the challenge we will face during the next couple of years.
We have some ideas on how to solve this, but how are YOU going to tackle this ?
Chat with us
Join us on October 13 to discuss this. Add meeting to your calendar