LLMs in prod: are we replacing deterministic automation with trust-based systems?

Hi,

Lately I’m seeing teams automate core workflows by wiring business logic in prompts directly to hosted LLMs like Claude or GPT.

Example I’ve seen in practice:
a developer says in chat that a container image is ready, the LLM decides it’s safe to deploy, generates a pipeline with parameters, and triggers it. No CI guardrails, no policy checks, just “the model followed the procedure”.

This makes me uneasy for a few reasons:

• Vendor lock-in at the reasoning/decision layer, not just APIs

• Leakage of operational knowledge via prompts and context

• Loss of determinism: no clear audit trail, replayability, or hard safety boundaries


I’m not anti-LLM. I see real value in summarization, explanation, anomaly detection, and operator assistance. But delegating state-changing decisions feels like a different class of risk.

Has anyone else run into this tension?

• Are you keeping LLMs assistive-only?

• Do you allow them to mutate state, and if so, how do you enforce guardrails?

• How are you thinking about this from an architecture / ops perspective?

Curious to hear how others are handling this long-term.

https://redd.it/1pt3xw5
@r_devops

Teleport!

I recently did a POC on Teleport as an intern, mainly around Kubernetes access, databases, and auditing. It feels like a pretty powerful “all-in-one” access layer, so I’m curious about real-world usage beyond the obvious basics. For folks using Teleport in production—what’s the most interesting or non-obvious use case you’ve implemented , I’d love to hear scenarios that are practical from devops engineer POV

https://redd.it/1pt2enr
@r_devops

AWS IAM for Startup Teams: Autonomy Without Chaos

We had developers blocked on infra emails for basic AWS provisioning because no one trusted IAM permissions.

I wrote about how we moved from “infra as a bottleneck” to developer autonomy using permission boundaries, without handing out admin access.

Would love feedback from folks who’ve solved (or struggled with) this in their orgs.

Link : https://medium.com/aws-in-plain-english/how-i-designed-an-aws-permissions-model-that-gave-developers-autonomy-without-losing-control-d50d03ca2a1d?sk=3d1d0ad4b5e3eb2c8a94cdb41f7f6a65

https://redd.it/1pt7gfk
@r_devops

First experience

Hello :D,
I've been in my first DevOps role for 3 months now, and I wanted to ask: what was your first experience like?

I used to be a developer with 2 years of experience, and I’m curious about how it felt for you when you started.

Right now I honestly feel really bad at it—I make a lot of silly mistakes and I’m starting to get discouraged. How did things go for you in the beginning?

https://redd.it/1pt9ug6
@r_devops

Suggestions on training.

Hi,

I've worked as a sysadmin for the past 15 years, always in the Linux world, initially with Red Hat and more recently with the Debian family. I've learned the main parts of AWS, GCP, and Terraform, and I also have recent experience with Git and GitHub (actions - CI/CD). I have an intermediate understanding of Python and networking.

The project I was working on has ended, and I'd like to hear your suggestions on what I should study to stay current.

https://redd.it/1ptau5o
@r_devops

Automations inside mid-size DevOps for non technical users

Hey everyone,

I’ve talked to a lot of non technical people working within DevOps teams, especially at smaller companies, and I keep seeing the same pain points come up when it comes to automating workflows:

Tools like zapier or n8n are tough to maintain. If someone builds a workflow and then leaves the team, it turns into a black box, especially for teammates without a technical background.

A lot of automation lives outside the team’s main communication tools like slack or teams, which makes it feel disconnected and awkward to trigger or adjust in context.

There’s usually very little visibility into what an automation is actually doing unless you dig into it, which makes trust and debugging harder.

We’ve been working on something in this area that focuses on natural language driven, context aware automations that live directly inside tools like slack, discord, or google teams so even non technical users can trigger, review, and tweak automations from where they already work.

I’m still trying to gather more feedback and get some opinions:

What’s been your experience with automation tools in small or mid-size DevOps teams?

What’s worked well, and what hasn’t?

https://redd.it/1ptc6gh
@r_devops

I built khaos - a Kafka traffic simulator for testing, learning, and chaos engineering

Just open-sourced a CLI tool I've been working on. It spins up a local Kafka cluster and generates realistic traffic from YAML configs.

Built it because I was tired of writing throwaway producer/consumer noscripts every time I needed to test something.

It can simulate:

\- Consumer lag buildup

\- Hot partitions (skewed keys)

\- Broker failures and rebalances

\- Backpressure scenarios

Also works against external clusters with SASL/SSL if you need that.

Repo: https://github.com/aleksandarskrbic/khaos

What Kafka testing scenarios do you wish existed?

\---

Install instructions are in the README.

https://redd.it/1pte4o9
@r_devops

How do I not waste my time in school?

I am a network engineer working in consulting by trade. I was fortunate enough to get into this position but as time is going on I'd like to be on the platform engineering side of things as I want to build other systems besides network infrastructure.

Now I know I can't just snap my fingers and hop so I am pursuing my bachelors at 28 in software engineering (specifically with WGUs BS and MS program - I am specifically going to shoot for their masters in dev ops program once I finish my bachelor's), I am happy to be able to finally be in a spot of life I can finally earn a degree.

What can I do to appropriately spend my time while in school to be in the best position to earn at least a junior platform engineer position. I'm pretty unsure about how to go about building a portfolio, connecting with people already in devops, and any other extra curriculars I can leverage to get me in. I appreciate any insight you folks might have or your guys experience in getting into the field.

https://redd.it/1ptf665
@r_devops

Traditional devops experience thought

So I don't use cloud as a primary part of my job. I do use it occasionally as a tool. I do an astronomical amount of automation for build and deploy. I am about to spend about 8 months standing up a front end in front of my automation to make a centralized signing and deployment much more user friendly


However I do feel like my career at this current company is on the sunset as I just don't really have much passion for mobile applications and there isn't a lot of space for me to grow into anything else and the depth at which I have to already be an expert is a lot further than I wanted to go


Problem is I don't have a lot of kubernetes experience. So I was thinking about creating a portfolio website that is essentially just a website that monitors its own infrastructure and is a visual representation of the automation

However I don't know if that's a worthwhile practice. I've had a hard time getting interviews lately even though I am a significant contributor at my current company which is in the fortune 200 list


I know that the hiring landscape is kind of bad right now and I honestly don't know if a personal project would even help me get hired as it seems like I'm competing with thousands of people that have the traditional devops experience


But I can do everything from mobile application architecture, I can stand up a web app on a small scale, I've been on the governance board for AI adoption in medical applications, and I have completely reworked a really old mobile application pipeline. When I first came to this company they had 400 bash Scripts and over 10,000 lines of code they handled all of their mobile application signing. The guy who wrote the system intentionally did not document it so that insured his employment

In the last 2 years I have fully documented the process and became a subject matter expert in my own right for mobile application signing and deployment. I've entirely Rewritten his tool to move off of Jenkins and on to git lab and positioned it to be deployed into the cloud if that was ever necessary

I have also trained an entire team of business analysts to handle every aspect of the mobile release process that isn't technical. I feel like I have overcome a lot and I feel like my resume doesn't do me a lot of Justice and because I was so pigeonholed into this shit hole of a team that is now amazing I've kind of stunted my growth


Like I could develop an architect Solutions like this on a whim very easily but at the same time nobody's going to let me touch their hybrid infrastructure because I don't have enough experience in the cloud. I don't know if you guys have any advice

https://redd.it/1ptdb0o
@r_devops

Anyone using Linear? I've got a couple 1-year coupons lying around.

I ended up with a few unused Linear 1 year credits from a deal I got earlier this month. I don't need all of them anymore, and they'll expire soon, so l figured I'd Give them on to people who want to improve their project + task workflow.

Linear really streamlined my planning + daily workflow. Instead of letting the credits expire, la rather give them to people who will actually use them to stay organized and ship faster.

If you want one, just comment "interested" or DM me and l'il send details.

https://redd.it/1ptekn0
@r_devops

I am building a Kubernetes operator dashboard as a personal project and having a lot of fun with it

Hi everyone,

I wanted to share a personal project I have been really enjoying working on.

Lynq is a Kubernetes operator that I am building on my own. While operating it, I kept running into a familiar DevOps problem. Once an operator is deployed, understanding what it is actually doing becomes harder than expected.

You can check pod status and logs, but questions like which resources are being managed, how they are connected, and what state the operator thinks they are in are not easy to answer quickly.

So I started building a small dashboard focused on operators.

The idea is to make day to day operator operations a bit more pleasant by:

* Showing relationships between operator managed resources
* Making current state and behavior easier to grasp
* Reducing the need to constantly jump between kubectl commands and logs

This is still early stage and not widely used at all. It is mostly a personal project, but I am excited about how it is shaping up and wanted to share it with the DevOps community.

I wrote a short blog post with screenshots and more details here: [https://lynq.sh/blog/introducing-lynq-dashboard](https://lynq.sh/blog/introducing-lynq-dashboard)

I would love to hear how others operate and debug their Kubernetes operators, and what kind of visibility you wish you had.

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

How to reduce api management costs for enterprise?

Our api management costs are getting out of control. We're spending way too much across apigee licensing, aws data transfer, and the team maintaining it all. We have around 200 apis serving internal teams and external partners, traffic is maybe 500M calls per month not massive but not small either.

The biggest cost drivers seem to be: apigee license, data transfer between regions, paying a vendor for ddos protection and three people spending 30% of their time just keeping it running

I looked at moving to aws api gateway but the per request pricing would actually cost us more at our volume azure apim has similar issues.

Anyone has managed to reduce these costs significantly without sacrificing reliability or features. Different vendors that are less expensive at scale? better ways to handle cross region traffic

I’m not looking to cheap out on something critical but this feels excessive for what we're getting, would love to hear what are you all doing.

https://redd.it/1ptjl69
@r_devops

AI makes coding insanely fast.


Right up until you run the thing.

Then it’s a wall of errors from code you didn’t fully write or fully understand, because half the logic was confidently invented. What was supposed to be a quick noscript turns into hours of debugging, refactoring, and figuring out why a small change nuked your environment.

Tools like ChatGPT help with speed, and I’ve found things like Cosine useful for tracing through larger codebases and understanding how pieces are wired together but none of it replaces actually knowing what the system is doing.

AI saves time on typing. It doesn’t save you from thinking.

https://redd.it/1ptmj51
@r_devops

https://github.com/LOLA0786/Intent-Engine-Api

I’ve been working on a small API after noticing a pattern in agentic AI systems:

AI agents can trigger actions (messages, workflows, approvals), but they often act without knowing whether there’s **real human intent or demand** behind those actions.

**Intent Engine** is an API that lets AI systems check for live human intent before acting.

**How it works:**

* Human intent is ingested into the system
* AI agents call `/verify-intent` before acting
* If intent exists → action allowed
* If not → action blocked

Example response:

{
"allowed": true,
"intent_score": 0.95,
"reason": "Live human intent detected"
}


The goal is not to add heavy human-in-the-loop workflows, but to provide a lightweight signal that helps avoid meaningless or spammy AI actions.

The API is simple (no LLM calls on verification), and it’s currently early access.

Repo + docs:
[https://github.com/LOLA0786/Intent-Engine-Api](https://github.com/LOLA0786/Intent-Engine-Api)

Happy to answer questions or hear where this would / wouldn’t be useful.

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

Best IaC platforms?


I am evaluating a few IaC platforms to sit on top of Terraform/OpenTofu for a multi‑cloud setup (AWS + Azure, possibly GCP later). The key technical requirement we have rn is to have a central layer for policy‑as‑code and guardrails across clouds, with drift detection that can raise PRs for remediation and a self‑service flow where app teams request environments through Terraform modules without editing raw HCL directly. One other big consideration for me is avoiding unnecessary abstraction. Ideally and if possible, the platform should have easy onboarding, simple integration with cloud providers and VCS, and not introduce overly complex access/auth models or identity layers that drive up overhead. I’m looking for something that enhances IaC workflows without becoming another system I have to maintain.

Right now I am looking at some of these options:

Firefly: Multi‑cloud platform with inventory and codification with Guardrails, policy‑as‑code, and drift remediation that opens PRs

Spacelift: Terraform/OpenTofu automation tool with flexible pipelines, strong VCS/CI integration, and policy hooks

env0: Platform with seemingly more emphasis on environment management, cost controls, and approvals around Terraform workspaces and modules

If you have experience using any of these for multi‑cloud governance, self‑service environments, etc., how well did they handle these things?

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

Best Terraform Cloud Alternative?

looking for a Terraform Cloud alternative for large team using multi‑cloud setup. We manage a few hundred workspaces across AWS and Azure with remote state, policy checks, and cost visibility wired into CI, but Terraform Cloud pricing and org limits are becoming an issue. What are people using instead to handle workspace orchestration, state storage, drift detection, and policy enforcement at this scale, preferably with SSO and audit logs built in?

https://redd.it/1pto5h1
@r_devops

How does adding monitoring/alerts process looks like in your place

I am trying to understand how SMB's are handling their Grafana / Datadog / Groundcover
dashboards, panels, alerts at scale.

furthermore, I try to understand how goes the "what should I monitor", "on what should be alert and at which treshold?"

how this process goes in your company?

is it:
1. having an incident
2. understanding which metric/alert was missing in order to detect earlier/prevent
3. add this metric, add the dashboard/panel and an alert?

is it also:
1. map on a regular basis (monthly) your current "production" infra/services/3rd parties
2. understand consequences, and create relevant alerts both app and infra?



wish to shed some light on it in order to streamline this process where I work

https://redd.it/1ptq8fl
@r_devops

Do you use paid tools for API testing?

We have been using Postman's free plan for API testing for a long time but we feel that it has become quite restrictive with limits on the number of users, collection runs etc.
I want to understand if it's worth upgrading to their paid plan or moving to some other tool?

View Poll

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

I want out

Maybe a grass is greener on the other side issue. But I’m so tired of being treated as a drain on the company.

It’s the classic, everything’s working, why do we need you, something broke it’s your fault.
Then there’s the additional why is your work taking you so long.

Gee maybe it’s because every engineer wants improvements but that’s not their job, that’s OPS work. Give it to one of the 3 OPS engineers.

So what can I do? Is there a lateral shift that would let me try and maintain a similar 150-200k salary range?

I hated school. Like I’ll suffer if that’s what’s required. But I’d prefer not. Maybe sales for a SAAS company? Or recruitment? I just want to be treated like an asset man.

https://redd.it/1pty4ok
@r_devops

Luxury Yacht, a Kubernetes management app

Hello, all. Luxury Yacht is a desktop app for managing Kubernetes clusters that I've been working on for the past few months. It's available for macOS, Windows, and Linux. It's built with Wails v2. Huge thanks to Lea Anthony for that awesome project. Can't wait for Wails v3.

This originally started as a personal project that I didn't intend to release. I know there are a number of other good apps in this space, but none of them work quite the way I want them to, so I decided to build one. Along the way it got good enough that I thought others might enjoy using it.

Luxury Yacht is FOSS, and I have no intention of ever charging money for it. It's been a labor of love, a great learning opportunity, and an attempt to try to give something back to the FOSS community that has given me so much.

If you want to get a sense of what it can do without downloading and installing it, read the primer. Or, head to the Releases page to download the latest release.

Oh, a quick note about the name. I wanted something that was fun and invoked the nautical theme of Kubernetes, but I didn't want yet another "K" name. A conversation with a friend led me to the name "Luxury Yacht", and I warmed up to it pretty quickly. It's goofy but I like it. Plus, it has a Monty Python connection, which makes me happy.

https://redd.it/1pu1o6t
@r_devops

github-ci: Lint your GitHub Actions workflows and auto-upgrade to latest versions

https://github.com/reugn/github-ci



I've been spending time managing GitHub Actions workflows manually across different projects. I built this tool to automate some of that and make it less tedious. If you find it useful, let me know - I'm planning to add more features over time, so contributions are welcome.

https://redd.it/1pu3beq
@r_devops