AI as a DevOps Collaborator: How We're Moving Faster Across Projects
We've been rethinking how our team approaches DevOps work by giving AI scoped, intentional access to our environments — and the results have been significant.
We’ve been tightening up workflows across all of our active projects lately, especially around AWS and AI-assisted development. What started as an experiment in process improvement has turned into a genuine shift in how our team operates.
The short version: we started treating AI like an actual member of the team — and it changed how we ship.
The Setup: Scoped Access by Design
Across projects, we’ve settled on a consistent pattern we call the “AWS CLI User” skill. For each project, we create a dedicated IAM user through the AWS console, scope its permissions appropriately, store API keys locally, and activate the profile with a single command:
export AWS_PROFILE=projectname-cli-devThat’s the entire setup. It’s intentionally minimal.
What that simplicity unlocks, though, is significant. Once a profile is active, Claude isn’t just answering questions in the abstract — it’s operating as a DevOps engineer with real context from our actual environment. It can inspect live resources, surface relevant details, and help us think through decisions against what’s actually deployed rather than what we assume is deployed.
This is a meaningful difference. Generic AI responses are useful. Responses grounded in the real state of your infrastructure are much more useful.
Teams Move Faster When AI Has Context
The reason this pattern has stuck with us is how much it compresses the feedback loop for the whole team.
Before, DevOps work often meant one person going deep on a task — digging through docs, piecing together configurations, verifying assumptions — while everyone else waited. That bottleneck is hard to parallelize because so much of the knowledge lives in one person’s head at any given moment.
When AI has scoped access to your environment, that knowledge becomes shared infrastructure. Anyone on the team can pick up a DevOps task, work through it with Claude, and produce an output that’s grounded in reality rather than guesswork. The experienced engineers don’t become less valuable — they become multiplied, because the baseline for everyone else rises.
A Real Example: Moving from Mailgun to Amazon SES
This played out clearly when we decided to evaluate our transactional email setup on one of our projects.
Mailgun is excellent for getting started. The API is developer-friendly, setup is fast, and the free tier is generous enough to carry you through early-stage work. But as volumes grow, costs scale quickly, and we wanted to understand our options before that became a constraint.
Amazon Simple Email Service was the obvious candidate for comparison. The economics are hard to argue with — especially for teams already running on AWS. The free tier covers a significant email volume when sending from EC2, and beyond that, the per-email cost is a fraction of most third-party providers.
We knew that going in. What we didn’t expect was how smoothly the migration would go.
Claude handled the full arc of it: comparing the services against our actual usage patterns, walking through the SES configuration, helping set up credentials, and validating the end-to-end flow. What would typically have been a half-day of context switching across docs, dashboards, and console configurations compressed into a focused working session.
The migration shipped the same day we decided to do it.
That kind of velocity isn’t about cutting corners — the work was done carefully. It’s about removing the friction that normally accumulates between “we know what we want to do” and “it’s done.”
What This Model Actually Changes
The traditional DevOps workflow has a lot of latency built into it. Decisions wait on availability. Implementations wait on expertise. Reviews wait on context being transferred from one person to another.
Giving AI scoped, intentional access to your environment doesn’t eliminate those steps — it dramatically shortens them. The AI arrives in every conversation already knowing the shape of the system. You spend less time orienting and more time deciding.
For our team, this has shown up most clearly in the DevOps layer: infrastructure changes, service migrations, configuration work. These used to be tasks that required the right person to be available. Now they’re tasks that any member of the team can drive, with AI providing the depth that used to require a specialist.
We’re still the ones making the calls. The AI is still operating within clearly defined boundaries. But the team moves faster, decisions get made sooner, and nothing waits on one person’s availability to unblock it.
We’re continuing to refine what this looks like across our projects — and we’ll keep sharing what we learn.