Brady Holt

Eduardo Ordax

Amazon Web Services (AWS) • 242K followers

Vibe coders will never understand this 😅 "Why would you need a 49" curved monitor just to write code?" Because that’s the only way to see the entire Java class name on the same screen... at the same time. This isn’t a setup. It’s a survival guide.

Vibe coders will never understand this 😅 "Why would you need a 49" curved monitor just to write code?" Because that’s the only way to see the entire Java class name on the same screen... at the same time. This isn’t a setup. It’s a survival guide.

Matt Martin

State Farm • 14K followers

Been using EMR Serverless a good bit lately and here are some opinions: The Good - Quick startups; when I submit a job, its up and running in about 3 minutes or less - standard spark; does not have the extra wrapper stuff that Glue jobs have; this makes local dev and testing a lot easier and more portable - security is straight forward; start job as an IAM role; role then needs proper permissions to assume various database roles, read/writes to s3; no confusion there The Ugly - class path resolution in EMR serverless is terrible. If you want to have a folder structure such as: src/main |----etl |----data ...In the ETL and data sub folders, you have to add code at the top of each module for a sys path addition. - additionally, I spent probably 4 hours debugging an issue where it said it could not find module "ETL", which was a sub folder I had with some code. What fixed that? Adding an stub/empty __init__.py file. And I was not using classes because this was simple movement of data from point A to point B, no multiple instances, no state required. God only knows why EMR is too dumb to find these modules without that init stub. ...maybe I'm holding it wrong. Needless to say, what took me several hours in EMR would have taken me probably 20 minutes in #databricks notebooks. There is a very strong and clear reason why the masses are gravitating to the databricks notebook experience...it just works. ...and yea, you can call it a skill issue if you'd like 😄. It's 2025 and I should not have to burn countless cycles just to get code that I know works to run in an environment. That is an utter waste of time that can be better spent delivering real value. </endOfRant> #dataengineering #databricks #aws

Jack Pines

Infoblox • 2K followers

And yet Big Tech is betting in AI replacing software engineers in a big way. Microsoft alone is basing full-time employees’ reviews on how much they utilize AI. I’ve lived this before as IT staffing was shipped offshore to lower cost areas like India and Argentina, saying that senior roles won’t be moved. Then they were once those offshore junior staff became senior. No surprise. Then they did it again with software engineers. Now they’re doing it again but with AI. There’s one HUGE problem with that, aside from the constant gutting of higher paid careers in the U.S., and that is how do you expect to have senior engineers to monitor for and correct AI mistakes if you don’t invest in junior engineers to be the next generation of senior engineers? That is my mission as co-founder of Between2Pines Consulting, to make sure junior talent is fostered to become the next generation of senior engineers. If you are a small to medium sized business looking to augment your staffing, but don’t have the budget to train up junior engineers, reach out to me to see how I can provide you mentored junior talent who will provide you senior-level results.

Michael Sargent

Lamp Light Labs • 564 followers

My most recent contract wrapped up in March and I'm available for new engagements — immediately. Here's what I do: I parachute into broken situations and fix them. Pipelines that won't deploy. .NET applications throwing errors in production. SQL queries grinding everything to a halt. APIs that stopped talking to each other. Automation that was never built in the first place. Deep experience across defense, federal, financial, and healthcare systems means I've seen most of it before. Recent work: — Designed and built a Git-based source control workflow for custom JavaScript inside a SaaS loan origination platform — replacing a UI-dependent process with no change history, no peer review, and no automation testing — Delivered compliance automation at Blue Cross Blue Shield of SC in 6 weeks — half the team, half the time originally estimated — 15 years at BAE Systems on Army and Navy defense contracts — Chairman's Award for Innovation 6 times I'm available on a W2 contract basis through staffing agencies. Short-term fix-it projects are my sweet spot, though I'm open to ongoing arrangements as well. If you know someone who needs a senior engineer for a specific problem — I'd appreciate the introduction. #SoftwareEngineering #AzureDevOps #DotNet #Python #SQLServer #CSharp #ContractWork

Christian R. Vozar

Christian Vozar is a senior... • 1K followers

Some thinking on Evals. I think it’s dangerously close to becoming so overloaded as a term that it obscures more than it clarifies. When doing what they do, evals should function like traditional software tests. Prevent regressions, speed up iteration, and provide sanity checks before shipping. The limitations are structural and by definition, they only reflect the failure cases we already know about. In environments where models are non-deterministic, frequently updated, and increasingly agentic, real truth appears to only emerge in production. Proper monitoring, experimentation, and error analysis in live systems can provide ground truth signals that evals cannot capture, especially when personalization fragments the definition of "good" into millions of micro-contexts. Evals have a role, but the future is not eval-driven development (please, God) but production-informed development, where small sets of smoke tests are paired with rich, real-time monitoring to surface what test suites can’t predict.

Some thinking on Evals. I think it’s dangerously close to becoming so overloaded as a term that it obscures more than it clarifies. When doing what they do, evals should function like traditional software tests. Prevent regressions, speed up iteration, and provide sanity checks before shipping. The limitations are structural and by definition, they only reflect the failure cases we already know about. In environments where models are non-deterministic, frequently updated, and increasingly agentic, real truth appears to only emerge in production. Proper monitoring, experimentation, and error analysis in live systems can provide ground truth signals that evals cannot capture, especially when personalization fragments the definition of "good" into millions of micro-contexts. Evals have a role, but the future is not eval-driven development (please, God) but production-informed development, where small sets of smoke tests are paired with rich, real-time monitoring to surface what test suites can’t predict.

Dan Gray

ChronoDyne Systems, Inc. • 3K followers

I’m seeing this pattern now and it’s getting boring: Someone comments "AI wrote this" like they’ve uncovered a scandal. Meanwhile... those same people are using coding agents every day. And here’s the funny part: A coding agent isn’t magic. It’s not "intelligence". It’s literally: try → fail → retry → burn tokens → try again → maybe succeed That’s the workflow. Agents brute‐force their way through problems. They generate code, run it, break it, fix it, break it again, regenerate, patch, retry all while chewing through tokens like a woodchipper. So who’s really "AI‐written" here? The person who writes a post with a bit of AI assistance... or the person whose entire codebase is being hammered into shape by a loop that runs 40 attempts deep? The irony is wild. If you’re going to shout "AI wrote this!", at least admit you’re using the most AI‐dependent workflow of all: agents that keep trying until the tokens run out. #AIEverywhere #AIAugmented #AgentOps #CodingAgents #AIEngineering #SoftwareRealityCheck #ModernEngineering

Robert W. Woodcock

The Process MechanicTM • 2K followers

The 60-Second Morning Ritual That Prevented a 1ドル Million Surprise A dev team added code. Three months later, someone noticed their cloud microservice bill had doubled. Cost of the oversight? An extra 4,000ドル per month. The terrifying part? Their sister team managed 350,000ドル in monthly cloud storage. Same blind spot. Same 3-month delay. Do the math: That's a 1ドル Million surprise nobody would've caught. Here's what kept me up at night at Comcast: Teams had dashboards. AWS Cost Explorer? It shows you last month's damage, not today's bleeding. By the time you see that spike, the bill's already sent, the moneys’ already gone. That doubling of expenses? It went unnoticed until accounting flagged it - three months after the code was deployed. Why? Because showing yesterday's costs isn't the same as preventing tomorrow's surprises. We built something different: Working with a Principal Architect, we created hierarchical dashboards in DataDog using daily cost estimates that rolled up from service → environment → total view. The breakthrough wasn't the technology. It was the ritual: • Check the top dashboard every morning • Takes 60 seconds • Green? Move on with your day • Not green? Drill down to find which service, which environment First week results: • Manager spotted a spike within 4 hours (not 3 months) • Traced to a specific service in staging • Fixed before it hit production • Prevented a potential 6-figure mistake The real insight? Your teams don't need more dashboards. They need one dashboard they'll actually check. Make it so simple that skipping it feels wrong. So clear that problems scream for attention. So fast that there's no excuse. Because somewhere in your infrastructure, there's a cost doubling every month. The question is: Will you catch it tomorrow morning, or in three months? - - - - - - - - - - - - - - - - - - - - - - - - - - - - Hi, I’m Robert, A Senior Technical Program Manager who believes the best monitoring is the kind that becomes muscle memory. When visibility becomes ritual, surprises become rare. 🔧 💭 What expensive surprise is hiding in your metrics right now? ↗️ Share if you've ever found a costly blind spot months too late ⏰ What's your morning dashboard ritual? The Process MechanicTM books: https://lnkd.in/dV9TTmVc #CloudCost #DevOps #FinOps #CloudComputing #CostOptimization #TechLeadership #Monitoring #DataVisualization #EngineeringLeadership #AWS #Azure #CloudOps

The 60-Second Morning Ritual That Prevented a 1ドル Million Surprise A dev team added code. Three months later, someone noticed their cloud microservice bill had doubled. Cost of the oversight? An extra 4,000ドル per month. The terrifying part? Their sister team managed 350,000ドル in monthly cloud storage. Same blind spot. Same 3-month delay. Do the math: That's a 1ドル Million surprise nobody would've caught. Here's what kept me up at night at Comcast: Teams had dashboards. AWS Cost Explorer? It shows you last month's damage, not today's bleeding. By the time you see that spike, the bill's already sent, the moneys’ already gone. That doubling of expenses? It went unnoticed until accounting flagged it - three months after the code was deployed. Why? Because showing yesterday's costs isn't the same as preventing tomorrow's surprises. We built something different: Working with a Principal Architect, we created hierarchical dashboards in DataDog using daily cost estimates that rolled up from service → environment → total view. The breakthrough wasn't the technology. It was the ritual: • Check the top dashboard every morning • Takes 60 seconds • Green? Move on with your day • Not green? Drill down to find which service, which environment First week results: • Manager spotted a spike within 4 hours (not 3 months) • Traced to a specific service in staging • Fixed before it hit production • Prevented a potential 6-figure mistake The real insight? Your teams don't need more dashboards. They need one dashboard they'll actually check. Make it so simple that skipping it feels wrong. So clear that problems scream for attention. So fast that there's no excuse. Because somewhere in your infrastructure, there's a cost doubling every month. The question is: Will you catch it tomorrow morning, or in three months? - - - - - - - - - - - - - - - - - - - - - - - - - - - - Hi, I’m Robert, A Senior Technical Program Manager who believes the best monitoring is the kind that becomes muscle memory. When visibility becomes ritual, surprises become rare. 🔧 💭 What expensive surprise is hiding in your metrics right now? ↗️ Share if you've ever found a costly blind spot months too late ⏰ What's your morning dashboard ritual? The Process MechanicTM books: https://lnkd.in/dV9TTmVc #CloudCost #DevOps #FinOps #CloudComputing #CostOptimization #TechLeadership #Monitoring #DataVisualization #EngineeringLeadership #AWS #Azure #CloudOps

Michelle Martin

My Ideal Recruiter • 5K followers

Stop Filtering Out Your Best Engineers Still posting job descriptions like: "BS in Computer Science required" "Must have 7+ years of React" "Looking for a ‘rockstar’ coder" You’re repelling top-tier devs who: Built real products without a degree Moved fast with new stacks Care more about outcomes than buzzwords Great tech recruiting starts with great filters. Not outdated ones.

Saul Jaramillo

MoonPay • 2K followers

Anthropic just dropped Claude Code Review → multi-agent swarm auto-scans every PR for bugs, verifies findings, ranks severity. Internal stat: substantive reviews jumped 16% → 54%, catches ~7.5 real issues on big PRs. 15ドル–25/review beta for Team/Ent. Game changer or expensive lint? 👀 https://lnkd.in/eVpR_vyD

Anthropic just dropped Claude Code Review → multi-agent swarm auto-scans every PR for bugs, verifies findings, ranks severity. Internal stat: substantive reviews jumped 16% → 54%, catches ~7.5 real issues on big PRs. 15ドル–25/review beta for Team/Ent. Game changer or expensive lint? 👀 https://lnkd.in/eVpR_vyD

Christopher Schmidt

innoQ • 838 followers

If you want to use a coding agent like Claude Code in an okayish isolation with nested Docker without privilege flag, this could give a clue: docker run --rm -it --runtime=sysbox-runc \ -v ${PWD}:/workspace -w /workspace \ cruizba/ubuntu-dind Because we want our agents to run in containers and execute and test their created code in containers, right? https://lnkd.in/dWfjzwYk https://lnkd.in/dNCDkW4G

Harshith Vaddiparthy

VentureBeat • 8K followers

My OpenClaw running on Opus 4.6 says: Expected ',' or '}' after property value in JSON at position 86 (line 1 column 87) __ what does that mean? how do I fix it?