Infrastructure Automation and the Promise of Autonomous Agents

The advent of public cloud computing brought with it many exciting ideas and methodologies, including a shift from CapEx to OpEx, the fundamentals of compute, storage, and networking in a consumption utility model, and countless other tectonic shifts. 

From a technical perspective, I found Infrastructure as Code, configuration management, and API-driven infrastructure to be revolutionary, as they enabled dynamic workloads, scaling, and a repeatable “cattle not pets” approach to infrastructure.

Fast forward to the present and we’re in the advent of AI, discovering new patterns and ways to work.

Much of the focus in the AI zeitgeist has been on productivity, as generative AI is particularly effective at content creation, and therefore impactful for content marketers and developers. 

But, I think there is an area of opportunity that has been less explored that connects directly back to the public cloud. 

While we can clearly apply AI to help people do their jobs, is there an impact that it can more directly make on our applications and infrastructure?

Can AI have an impact on automating our applications and infrastructure?

IaC is fundamentally code, which we’ve already established is an area where generative AI can accelerate problem solving. With the ability to generate code, generative AI can potentially be leveraged to drive automation and decision making in a completely new way. 

AIOps has been a thing for a hot minute, but it's largely been focused on predictive models, where observability applications can predict disruptions to workloads before they happen. 

In spite of a lot of buzz about agentic AI, we’ve not really reached the tipping point for autonomous agents, but the potential for cloud infrastructure is quite interesting.

Imagine a scenario that many technologists encounter – POC development.

 Before embarking on a potentially costly and time-consuming engineering effort, I’ve often spent a smaller, fixed amount of time and effort to test the waters, experiment, and inform my decisions. 

These POCs generally take the form of test workloads with a variety of approaches to solve a problem with clear and measurable outputs that can be compared.

These test workloads would be developed, put into sandboxed environments, and then benchmarked against one another. 

While these experiments are a significantly smaller investment than an entire project, they still take time and money and sometimes lead to “escaped prototypes.”

If an agentic AI can be provided with a thesis, a sandbox, and a set of target metrics as input, it's not unreasonable to expect that it would be able to generate the code for a set of experiments as output and then hand off to autonomous agents to execute them and report back on the findings.

As a developer, I often find myself sucked into an addictive vortex of problem-solving, quickly iterating, testing, and asymptotically approaching a solution late into the night. 

Good luck getting to sleep if your brain wants answers as quickly as possible! I look forward to a world where I can go full Tony Stark and ask Jarvis to carry on experimenting while I hit the sack. Let’s just be sure we don’t end up creating Thanos. I’d prefer not to be snapped out of existence.