Built an AI DevOps assistant for AWS, NEED feedback..

Hey everyone,
My cofounder and I are building an AI-powered DevOps assistant aimed at startups and engineering teams using AWS. We'd love your raw, unfiltered feedback on the idea before we go further. 🙏

It’s basically a chat-based DevOps co-pilot that connects to your AWS account and helps you manage infra using natural language. It can:

Answer questions like:
“How many EC2s are running?”,
“Why are my costs high this month?”,
“Which stacks are failing?”

Convert prompts into AWS CLI commands (editable + safe approval flow)

Generate, iterate, and deploy CloudFormation templates from natural language

Integrate with GitHub/Bitbucket to:

-Scan repos for CloudFormation
-Trigger existing CI/CD pipelines
-Stream logs and diagnose failures
-Apply rule-based fixes via PRs

Enforce IAM-permissioned access, full audit logs, and org/team-based controls

We’re planning to add Terraform support next (already being requested).

☁️ This is why we’ve built it:

Infra is complex, DevOps is expensive, and a lot of startups struggle to operate AWS safely. We want this tool to feel like a senior DevOps engineer who answers questions, gives you the CLI/code to act, and handles pipelines safely with approvals.

https://redd.it/1q3dmd4
@r_devops

When is old?

At what age should someone hang their hat on trying to get in the door? What door should the older try for?

https://redd.it/1q3jlw7
@r_devops

One Windows package manager to rule them all?

Just came across a nice articsl about an unfair that brings all the various package managers together.

I personally mainly use chocolatey as it what integrated into the tool company use, however this one "UniGetUI" brings them all together into a gui.

I haven't tried it myself yet but the artical seems to good not to share.

https://www.makeuseof.com/replace-microsoft-store-with-unigetui-package-manager/

https://redd.it/1q3kln2
@r_devops

Many companies are moving towards Dev-owned DevOps.

I’m seeing a trend where companies want developers to handle DevOps work directly.

For someone working as a DevOps engineer, what’s the best way to adapt?

What new skills are worth learning, and what roles make sense in the future?

Curious to hear how others are handling this shift

https://redd.it/1q3h19o
@r_devops

CI/IaC is basically a control plane now… what guardrail helped the most?

It feels like everything is a control plane now. GitHub Actions, IaC pipelines, internal platforms, agents, all of it.

And the failure mode I keep seeing is “one small change lands everywhere” because the blast radius is huge and rollout/rollback isn’t really a thing.

Curious... What’s one guardrail you added that actually helped?

Canaries, progressive delivery, env isolation, policy checks, drift detection, JIT admin, whatever… doesn’t have to be fancy.

https://redd.it/1q3oifo
@r_devops

Is Kubernetes here to stay for a long time?

Is it worh investing time in learning K8s or it will be hidden under PaS? Is it a must have skill for every DevOps in the future or it is expected to be buried under other technologies?

https://redd.it/1q3qgdx
@r_devops

Sci-Fi Author needs your help - "End of Integers"

Hey folks! I'm a career IT Ops Engineer, and Author, with just enough programmatic knowledge to be dangerous. I'm writing a Sci-Fi novel, and need your advice.

It's the year 2711, and I have an android-like bot that works in a research lab. She has a malfunction when her human boss ask her a question that she isn't supposed to answer.

That causes an error that makes her verbalize the terms and conditions of the leasing contract that she's governed by. Not in an informational way, but one that shows she's had a failure and not acting right.

When she's done, there's a one-second pause, followed by the statement End of Integers, which she says like it's a punctuation mark.

EDIT - I want the answer to sound programmatic, but also vague and not possible.

My Dev wife thinks it's a brilliant idea, since there is no such thing as an "end of integers."

My thought is there's a safeguard to keep her from telling anyone what she knows, but the code for the safeguard has a flaw that makes her say End of Integers.

1. Keep this, or use another type of error?
2. If another, which one would make more sense, for what I need to accomplish?

Thank you, and may your Secrets Management never fail, and blow up your Sprint schedule :)

https://redd.it/1q3rjh6
@r_devops

UAT for 40 +

We are rolling out a chatbot for our organization. Leadership wants all of corp tech to be able to soft test the feature and provide feedback. Jira ID, Acceptance Criteria, Pass/ fail, stengths, weaknesses.

Normally i would have test steps but its really launch the bot and ask it questions related to denoscription/acceptance criteria.

My queation. How do you distribute and track something like this? I normally do feature releases which is done via email. This seems like it might be better on a Microsoft form with a power automate to a sharepoint list for metrics. Its 40 + scenarios though as well, add that to the problem on how to distribute and track question.

https://redd.it/1q3tzeb
@r_devops

Company I work for realized AI can’t replace DevOps and now Hiring again

Hi folks, I work as a freelance DevOps engineer, and in 2020–2022 I used to get 2-3 recruiter calls a day.. those were crazy times. It started to slowly fade off, and by mid-2023, although I still managed to get offers, it was noticeably harder.

Currently, the company I’m working at has a large proportion of developers compared to the DevOps team (I’d say \~15% DevOps, 85% devs). Our management tried multiple shiny tools to improve our processes, but we ended up using AI only for PR reviews and even that is mostly for pre-screening. We still have to manually review things since AI makes mistakes and hallucinates.

For past few years usual response around here was "Hey, these guys don’t know how to use AI and .. it’s a skill issue." but imo These folks haven’t dealt with complex infrastructure beyond boilerplate to think AI can automate DevOps.

During the past three years, I've heard all sorts of things: "Everything will be automated," "It’s just the first year of AI wait and see in a couple of years there won’t be dev jobs," "Devin will eliminate engineers.. (LOL to this one)", and so on. All this hype and bubble kept growing, yet where I worked there were no meaningful headcount reductions beyond cutting back on intern and junior roles doing mostly grunt work and boilerplate and even that ended up hurting us.

Anyway, all of this could have remained speculation, if not for the fact that DevOps positions previously considered redundant due to "more efficient processes" are now being filled again, and the 5-6 DevOps engineers on our team are so overworked that we urgently need to hire more people.

In short (TL;DR), I haven’t seen any meaningful AI automation beyond what we already had, nor did it add much real value to our team. At best, it made us slightly more efficient, but at the cost of reduced maintainability and more complexity in the codebase. If you enjoy working in DevOps, there are still plenty of opportunities out there and likely more going forward.

https://redd.it/1q3ugf8
@r_devops

Another Helm Chart for Garage (MinIO Alternative for Homelabs & Small Deployments)

After MinIO abandoned the open-source project, I needed a new S3-compatible object store for my homelab. I tried the usual suspects (SeaweedFS, Ceph, etc.), but Garage stood out for its simplicity and focus on small, geo-distributed clusters.

I have published a Helm chart that goes way beyond the official one, making Garage a drop-in replacement for MinIO with a much smoother experience for Kubernetes users.

Repo: https://github.com/datahub-local/garage-helm1

What makes this Helm chart better than the official one?

1. Automated cluster configuration: No more manual CLI or YAML hacks. Just set your layout, buckets, and keys in values.yaml or secrets and a job will set up them for you.
2. Built-in WebUI: Deploy the Garage WebUI with a single flag for easy management.
3. Gateway API support: Native support for Kubernetes Gateway API (plus Ingress), so you’re ready for modern K8s networking.
4. Grafana dashboard & ServiceMonitor: Get instant metrics and dashboards out of the box.
5. Extra resources: Inject any custom K8s manifest (Secrets, ConfigMaps, etc.) directly via values.yaml.

Big thanks to \#wittdennis — this chart is based on his original Helm chart for Garage!

If you’re looking for a MinIO alternative that’s actually open source and easy to run at home, give Garage (and this chart) a try. Feedback and PRs welcome!

https://redd.it/1q3utve
@r_devops

Those using GitLab + MS Teams - how do you handle MR notifications?

The native GitLab integration for Teams is pretty basic and Microsoft is retiring Office 365 connectors soon.


I've seen tools like PullNotifier for GitHub + Slack, but nothing similar for GitLab + Teams.


Anyone found a good solution for:

\- Getting notified when assigned to review

\- Avoiding channel spam from every commit/comment

\- Tracking which MRs are still waiting for review?


What's your workflow?

https://redd.it/1q3wxtu
@r_devops

What OS do you daily drive, and why?

I'm curious about people working in the field and why you use one OS over another?
Are there tools you've found that only avaliable on your distro of choice, is it because of stability, is it because of less bloat? Maybe it was the only option or you just like it?

https://redd.it/1q3zk3k
@r_devops

Eager to learn ,would love some structure

For the experienced DevOps engineers, if you were to go back to the beginning, what would you do to make sure you have the right skills for DevOps in today’s market?

I want to learn DevOps this year. I tried at the end of last year and I’d feel so discouraged looking at all the tools I am required to learn. I have seen some people say that “DevOps is a senior position job.”

I have an AWS CCP certificate and I have soo much time on my hands.

What advice would you guys give me?

https://redd.it/1q40oud
@r_devops

Experienced sysadmin cannot pass a coding interview. RIP

I'm an experienced sysadmin (15 years) looking for a job, and it looks like most companies are asking for coding skills now. The Leetcode challenges I've attempted do not mirror my experiences with Python at work, and I am banging my head against the "easy" ones.

I am 60% through "Python Data Structures & Algorithms + LEETCODE Exercises" on Udemy, and I still do not recognize the patterns that are presented in Leetcode problems.

Am I digging in the wrong direction here? How should I be studying? Should I switch careers at the age of 40 and become a toilet farmer?

https://redd.it/1q42yvj
@r_devops

Do you also struggle with non-prod environments being left running “just in case”?

Hi everyone,


I’m curious if this is a common issue or just something I’ve seen in a few teams.

In many companies I’ve observed, non-production environments (dev / test / staging) are often left running 24/7, even though they’re only actively used during working hours.

When I ask why they’re not shut down after hours, the most common answer is:
“Just in case we need it.”

Not because they’re actually needed at night, but because people are worried that:
- someone might suddenly need access
- shutting it down could cause problems
- no one wants to be responsible if something breaks

Does this sound familiar to you?

If yes:
- how do you currently deal with this?
- is it mostly a cost issue, a risk issue, or an ownership issue in your team?

just trying to understand how widespread this problem really is.


https://redd.it/1q413fo
@r_devops

Asked to spread into ML-Ops, but it's new territory. Being required to find related certs but unsure where to start.

I'm a DevOps engineer for a fortune 500 tech company. On my team, I'm the sole person in my role. Been here for 6 years. In fact, for my entire org, I'm only 1 of a handful of us. Our CICD pipeline is very solid and simple to maintain. Most of my work centers around DevSecOps instead of just DevOps. I KNOW that my company is paying me less than what I'm worth, but when the market is "iffy", I don't want to rock the boat. I do well in my role, but even 6 years later I still feel like there's a bit of imposter syndrome going on, despite consistently good recognition and reviews.

So I helped out on an AI-centric hackathon with work and provided all kinds of tech-related assistance to the different teams, such as provisioning new cloud products, creating DNS records for them, debugging various issues, things like that.

Afterwards, I'm now being told that for FY26, I have a personal goal of related certification to attain, but it's on me to find the relevant certs with which to get. I know what AI is. I can bust out a set of prompts that are rather decent. That's about the extent of it.

So as a DevOps Engineer, who acts as a consultant for his team on the more technical side of things, I feel it's my responsibility to not only be able to deploy various models, but also interact with various closed models, as well. And this includes Generative AI for text-based resources and image-based resources as the company I work for is one of the largest graphics-related companies in the world, apparently that's important.

So where do I start? I feel I need to know what's involved at a low level, hence the thought about deploying models. Beyond that, it's pretty new territory to me.

https://redd.it/1q44glk
@r_devops

I got tired of "shallow" GCP labs, so I built a soulful, production-ready scenario. Looking for technical feedback.

TL;DR: I created a GCP tutorial scenario as a pilot for a bigger series. It’s designed to read like an engaging article rather than dry documentation. I’m looking for feedback on the architecture and flow.



Hello,

After spending quite a bit of time on GCP designed labs (on CloudSkillsBoost) and courses I came to a conclusion that these either go in depth on very shallow scenarios or they skim over a lot of important stuff in more complex topics. The end status, I feel, is that you end up with this scattered knowledge about the platform that you then might struggle to put together into a secure, prod ready setup.

I decided to build a set of tutorials that don't just give you commands to copy, but explain the why. I’ve poured my personality into this - I wanted to make it an engaging "story" that you actually enjoy reading, rather than just checking boxes and copy pasting the commands.

Here is the TLDR about the scenario from the repository:

## TL;DR - what you'll learn and what we'll use
### GCP Services Used:
- Cloud Build (with Buildpacks)
- Cloud Run (backend)
- Cloud Functions (async processing)
- Pub/Sub
- Cloud SQL (Postgres)


### What you will learn
- How to deploy serverless applications to Cloud Run & Cloud Functions
- How to connect GCP-managed services to resources inside your own VPC (spoiler: it’s not as magical as marketing suggests)
- How to build a secure, end-to-end serverless microservice architecture
- How to apply Principle of Least Privilege (PoLP) to serverless components
- How to avoid Dockerfiles using Buildpacks, reducing ops overhead
- And finally how to tie this all together

I come to you, fellow engineers, to ask for feedback on the the technical accuracy, the flow, and the "engagement" factor. Does this feel like something a mid/senior dev would actually find valuable? My friends haven't been much help in the review department, so I'm reaching out to the community for some honest peer review.


Here's the link to the scenario:
https://github.com/brzezinskilukasz/gcp-tutorials/tree/main/scenarios/1

https://redd.it/1q42g5w
@r_devops

We built a GitHub Action that could have prevented the CrowdStrike outage. It's free.

On July 19, 2024, CrowdStrike pushed a config update that crashed 8.5 million Windows machines. The root cause: 21 fields validated against a 20-field schema. The unvalidated field caused a null pointer exception.

We ran that deployment profile through ARBITER:

Bad deployment:
0.335 null pointer exception ✓ ← RANKED FIRST
0.235 memory access violation ✓
0.149 safe execution ✓
0.120 system crash ✓

Good deployment:
0.257 safe execution ✓ ← RANKED FIRST
-0.068 null pointer exception ✗ ← REJECTED
-0.094 memory access violation ✗ ← REJECTED
-0.176 system crash ✗ ← REJECTED

ARBITER is a semantic coherence gate. It checks if your deployment profile coheres with "safe execution" or "failure modes" before you push.

Add it to your pipeline:

uses: arbiter-engine/arbiter-action@v1

Marketplace: https://github.com/marketplace/actions/arbiter-deployment-coherence-check

It's free. MIT licensed. 26MB deterministic engine.

Your move.

https://redd.it/1q4bzef
@r_devops

A small browser-only page I built for quick config diffs

Been working on a side project over the holidays. Built a small browser only page that lets me paste two configs and diff them locally. It flags changed values and a few things that tend to usually bite.

No accounts, no uploads, no backend. It just runs in the browser.

Hope it helps!

https://configsift.com

https://redd.it/1q4ctjt
@r_devops

ai made shipping faster but understanding slower



lately i’ve been thinking about how different building feels now compared to a few years ago. getting something off the ground is insanely fast. scaffolds, endpoints, ui, all done in a weekend. but when something breaks, i’m spending way more time reading than actually writing code.

i’ve ended up using different tools depending on what i’m working on. GitHub Copilot for in-editor autocomplete and quick suggestions, Replit Agent when i want help across bigger chunks of work, Claude Code when i need to talk through a codebase at a higher level. and on larger or messier repos, i’ve found cosine surprisingly useful to trace how logic flows across files when my mental map falls apart. it’s not doing magic, it just helps me see what already exists without burning energy.

it feels like the bottleneck shifted from “can i build this?” to “do i actually understand what’s already here?” curious how others are dealing with this. do you stick to one ai tool, or do you end up with a stack where each thing does one job well?

https://redd.it/1q4ermd
@r_devops