This past week, I got to do some more experimenting with MCP, and honestly the implications of that technology are much bigger than I expected with regard to legacy systems. It’s so exciting to get to investigate new technologies like this that really change the paradigm of your thinking about how you can solve problems with technology.
For me, as I look at the AI landscape, I’m seeing a “New” DevOps problem that we need to solve. There is no question we are, not even entering, in the beginning of the age of AI. And nothing has changed the world of software development this much since Object-Oriented Programming. But to me, I’m seeing a new devops challenge that needs to be solved. We have lot of brilliant engineers who are turning their attention to building new models, new agents and new ways of getting solutions built faster. But along with that speed comes a new challenge. Specifically, how we take those solutions and get them into the hands of users at the same speed. Now the reason I keep putting “new’ in quotes is because this isn’t a new problem, just the latest iteration of it.
You could make the argument that when containers hit the scene, there was this idea of “hey I can build the app in a container much faster, and we saw a challenge of “Yeah but containers on their own don’t scale.” And a bunch of new solutions were born out of that such as Kubernetes, Docker Swarm, Service Fabric, and many others. I would argue Kubernetes is the lone survivor of that in a lot of ways. But that period of turmoil was necessary to figure out how to solve the problem.
The next iteration of this in the data science world was the rise of the Jupyter Notebook. Here comes this brand-new technology that made it much easier to build out and do data science work. But once again, “Yeah but notebooks can’t be deployed in production?”
The classic answer in a lot of cases to these situations seems to always come back to “DevOps” and this idea of a “Devs that handle both sides of the equation.” But honestly, this is becoming more of a utopian dream that can’t work. The numbers of tools, technologies, patterns continue to explode along with the concerns. Making it impossible for one person to be the jack of all trades.
For me, the question I continue to bring up and engage in with customers is “Ok, God Forbid you are ridiculously successful, then what?” And as weird as this question is, the point I’m driving home is that when we are building these solutions to show to leadership or stakeholders to drive adoption. What happens when they save “I love it, what has to happen to get it in the hands of end users.” What we are seeing is that these vibe coded solutions, or AI solutions can run locally or in a lab environments. But the idea of getting them into a production setting is concept that is not being addressed until they get by in.
DevOps has shown us this is a mistake, and we really need to consider the idea of building these out in our mind and having plans on how to move to a scalable solution.