TL;DR — Key Takeaways
- Agentic coding could accelerate platform engineering adoption. OpenAI’s Aidan Cunniffe says lower development costs make it more practical to turn libraries, tests and other components into reusable services and AI skills.
- Shared capabilities could speed development and patching. Platform teams can manage resources that AI agents invoke across development teams, reducing repeated work.
- Oversight remains essential. Teams need controls around skill updates, independent validation of generated code and coordination among agents working in parallel.
The rise of artificial intelligence coding will ultimately lead to more “platformization” being applied to DevOps workflows now that the math used to assess costs has fundamentally changed.
Speaking at an AI DevCon event, Aidan Cunniffe, a member of the product staff for OpenAI, told conference attendees that it’s become much more practical for platform engineering teams to convert libraries, tests and artifacts into reusable services that can be invoked via agentic engineering workflows.
Historically, it would have been too complex to capture the artifacts that application developers used to construct an application. With the rise of AI skills, however, the cost of converting libraries and other primitives into AI skills that can be reused across teams of application developers has dropped to the point where it is now affordable for platform engineers to build and manage them, said Cunniffe.
Individual developers will still need the expertise required to coax AI agents to invoke those skills, but as more of them become adept at working with AI agents, the rate at which applications can be confidently built and deployed will dramatically accelerate, he added. It will also become simpler to deploy patches as needed at a time when AI models are also making it much simpler to discover and exploit vulnerabilities, noted Cunniffe.
Finally, organizations will also discover that platforms make it simpler for the average end user to also build and deploy applications, he added.
Of course, a significant amount of oversight will be needed for most platform engineering teams to realize that vision. AI agents, for example, should not be allowed to update AI skills, and platform engineering teams should take care to ensure that AI agents are not conducting tests on their own output that they will invariably pass, said Cunniffe.
Naturally, each platform engineering team will need to determine at what pace to deploy AI agents. Challenges range from finding ways to manage AI agents that are executing tasks in parallel to enabling end users to “vibe code” applications with the same rigor that professional developers apply to the building of software. Regardless of approach, simply adding AI agents to existing workflows is likely to create as many issues as it solves, especially as the number of potential vulnerabilities that might be discovered in that code increases.
Hopefully, it won’t be too long before existing workflows catch up with the rate at which application development teams are currently using AI coding tools to generate code. Much of that code is not being properly reviewed before being deployed in production environments simply because existing DevSecOps workflows are being overwhelmed. The challenge and the opportunity now is finding a way to apply best platform engineering practices at scale to ensure that, going forward, that code doesn’t wind up becoming more trouble than it might actually be worth. In the meantime, however, platform engineers should expect that at some point they will be called upon to re-engineer workflows that arguably were deeply flawed long before AI ever came on the scene.