About
I’m Debo, an infrastructure engineer-turned-founder with a career spent building and…
Activity
8K followers
-
Debosmit Ray (debo) posted thisA DevOps engineer spent 2 quarters setting up Kubernetes. Our engineers would probably take 1 day. Maybe I should give my engineers 179 days off every 6 months? If your problem was real, you wouldn't have had 6 months to spend on it. Real business problems don't wait. They light your hair on fire, and you put it out before it burns everything down. If you have infinite time to solve something, the problem probably wasn't that big to begin with. 6 months in our industry is basically forever. The engineers I respect most get something done in 1 day that other teams think will take two quarters. They're solving something real. 179 days. For a Kubernetes setup? Hire better.
-
Debosmit Ray (debo) posted thisThe cost of generating code just hit 0. The cost of thinking has never been higher. Over the last year, we've been seeing something consistent across nearly every Kubernetes environment we analyze: More deployments. More configs. More services running in clusters. Millions in waste. It's not a coincidence. We're now living in a world where predicting the next token costs almost nothing. Code generation is essentially free. Engineers are shipping more, deploying more, iterating more. The volume of changes hitting production Kubernetes clusters has gone up dramatically. But the quality of thinking behind those configs? That hasn't kept up. Here's how it usually goes: An engineer needs a Kubernetes config. They find a tutorial. The tutorial is written to be simple; that's the whole point of a tutorial. No tutorial will ever say "When you actually run this at scale, don't do these things." It'll have a tiny disclaimer at best. So the engineer copies it. Then they use an AI tool to fill in the gaps. Then they use the same AI tool to review what it just wrote. You don't get the fox to guard the hen house. But that's exactly what's happening, at scale, across the industry right now. The result isn't broken clusters. The result is clusters that work fine (sometimes even great) but cost a fortune. Jumping up and down isn't moving forward. Kubernetes doesn't care how fast you're shipping; it'll charge you for every idle CPU cycle either way. This is what we're actually seeing in clusters today: → Demo configs running on real infrastructure. Copied from tutorials built for simplicity, not production. → Overprovisioning that compounds and becomes expensive. More deployments means more churn in resource allocation. Autoscalers can't keep up. → Your cluster pays for the peak. AI-generated code that has no idea if your function will serve one user or five million. It writes for the average case. Kubernetes was built to manage containerized workloads at scale. It wasn't built to think. That part was always supposed to be on us.
-
Debosmit Ray (debo) shared thisBy the time a platform engineer is urgently asking Claude for "Help with Kubernetes cost optimization," it's already too late. We go to KubeCon to meet them before they think to ask AI for help. We're a Kubernetes optimization company. KubeCon is where every single person who runs Kubernetes inside a real company shows up; the people actually responsible for keeping clusters alive, day to day, week over week. These are the platform engineers who set up the software. They're the ones who know, somewhere in the back of their head, that waste is killing them. And they're the ones who get the polite but annoying memo: "Our infra spend is too high. We need you to figure it out." We want to already exist in their head when that conversation happens. We go to be known. That's it. I was at Uber long enough to know how big (read: stupidly obvious) decisions actually get made. No one wakes up and decides to fix their cloud waste on a Monday morning. It starts with a conversation. Then someone mentions a name they've heard before. Then 2 weeks later someone remembers to follow up. We're going to every KubeCon this year. Because the best time to be known is before anyone needs you. If you're in Amsterdam this week, stop by our booth and get to know the DevZero team.
-
Debosmit Ray (debo) posted thisI watched Uber go from 50 engineers to 1000s. From monolith to one of the most complex microservice architectures in the world. But now at DevZero we run a monolith. I want to hate myself for it, but I can't. Because it's working. It was a deliberate choice. Monolith vs. microservices isn't a binary decision you make once and live with forever. It's a spectrum, and where you sit on it should be driven by what your team can actually operate, not by what sounds cool. I've been in the room when Uber made those calls. They weren't made because microservices sounded impressive. They were made because the business demanded it and the team had the scars that proved they would actually follow through. Right now we run different configurations of the same monolith. It's not simple, but it's ours. Every engineer on our team knows exactly how the whole system fits together. Wake any of us up at 3am and we'll give you a detailed rundown. But that's the part people skip when they read the "we migrated to microservices" blog posts. They see the architecture diagram. They don't see the years of muscle-building that made it possible to operate without everything catching fire. When people jump to microservices without that foundation, they don't have the deployment pipelines. They don't have the observability. They don't have the on-call culture. What happens is that they’ll end up with 20 services, no tracing, and engineers who only understand their one corner of the system. Then someone gets paged at 3am because nobody knows where to start. Complexity without capability is just debt you haven't paid yet.
-
Debosmit Ray (debo) shared thisManager: "We need to migrate to Kubernetes." Engineer: "Okay, I'll put together a team." Manager: "How long will it take?" Engineer: "6 months, maybe more. We'll need 6 people on it." Manager: "That works for us." Me as a fly on the wall: Shaking my head and losing my freakin’ mind. 6 months is an eternity. And 6 people? Are you kidding me? I come from Uber, where you could staff 20 engineers on a single infra problem. And sometimes that made sense. But mostly it meant 80% of the time went into meetings, alignment, and internal communication instead of actually fixing anything. The engineers who moved mountains were the ones in every outage. The ones who understood how the whole system fit together, and the people you relied on to get something done in 10 minutes that a committee would spend 2 quarters debating. Every company has maybe 2 or 3 of these change agents (if they're lucky). Ignas Mikalajunas is one of ours. And because of people like Ignas, we’ve learned that the only thing that actually works is finding the 1 engineer who gives a damn, giving them the access they need, and getting out of their way.
-
Debosmit Ray (debo) shared this8 months ago, my co-founder and I sat across from our investors. I was sweating through my shirt and shaking so much you could hear my teeth chatter. We were about to tell them why the thing they funded no longer existed. And I was scared they were going to set me on fire. They didn't. Thankfully, they were intrigued and stuck with me. See, the original DevZero was a developer productivity platform. I believed in it. The team believed in it. My investors believed in it. But our customers were using us for something else entirely: Managing their infrastructure more efficiently, in ways we hadn't designed it for or even considered. The writing was on the wall. So...we made the switch to become a K8s optimization platform, despite it being a full left turn from the original vision. In a startup, you hypothesize. You test. You learn. You move. And most of the time, you fail. But when you stumble on something that works… When you are actually fixing real problems that real people have… When your clients are literally begging you for more… You drop everything to do more of whatever that is. Big companies plan, build consensus, and move carefully. But following the same SOP of a massive company like Uber would have killed DevZero. We stayed agile, and it saved us. We never sat in a committee meeting or sent 3000 emails. We just noticed what was working and went for it. That conversation with my investors still sits heavy. But looking back, I'm glad it does. It was that struggle that made us who we are.
-
Debosmit Ray (debo) shared thisI’ll be at KubeCon in Amsterdam in 2 weeks! If you’re running Kubernetes at scale, chances are there’s still room to improve how efficiently your clusters are using compute. We can show you how we’re helping teams reduce Kubernetes costs by around 30% and help you do the same, or we pay €10k.* If you’re attending KubeCon and want to compare notes, feel free to grab some time here: https://lnkd.in/gHHAU-HR *𝘊𝘰𝘯𝘥𝘪𝘵𝘪𝘰𝘯𝘴 𝘢𝘱𝘱𝘭𝘺.
-
Debosmit Ray (debo) posted thisEngineering Leaders: Kubernetes doesn’t kill startups, your poor decision-making does. When you come back from a conference ready to adopt K8s because of your need to feel useful, engineers are stuck implementing your bad ideas. This is a typical story I hear on repeat: 1. The VP of Engineering hires a DevOps engineer to "modernize their infrastructure." 2. 6 months later, you have a malfunctioning cluster, a migration postmortem that won't end, and half of your runway gone. 3. Then you write a blog post blaming Kubernetes. My heart doesn't bleed for you. A tool is just a thing to solve a problem. When it fails, the failure is almost never the tool... It's the manager who couldn't evaluate the real reason for the hire. It's the team that can't execute because they don't understand the tool. It's the 6-month timeline that should have been your first signal that this wasn’t a good idea. Real business problems don't wait 6 months while you figure out YAML. Stop blaming Kubernetes. Hire better. Decide better. In that order.
-
Debosmit Ray (debo) posted thisYour cloud bill spiked last month. With your head on fire, you grab the fastest fix; a savings plan. The invoice drops 30%. You pat yourself on the back. But you didn’t fix anything. Savings plans are like a Band-Aid with a cute puppy on it; a pretty solution that doesn’t actually solve anything. The cloud loves this by the way. You just handed them guaranteed monthly revenue. And in exchange, they gave you a discount on the waste you were already paying for. Pat yourself on the back again and get yourself a juice box this time. Your clusters are still running at 15% utilization (on a good day…) You're still requesting 100 and barely using 10. The savings plan didn't touch that. It just made it feel manageable. Discounts change the price. Not the waste. If you actually want to stop the bleeding: → Pull your top workloads by requested CPU → Compare requests to actual usage over 7–14 days → Find the biggest gaps: the "100 requested, 5 used" cases → Right-size with a rollback plan, track error rates, and tweak → Check at 24h, 48h, and adjust based on what you’re actually seeing This is much less glamorous than a phone call to your AWS rep. But it's the only thing that actually works. If your cloud bill is on fire, read more about why a savings plan won't save you: https://lnkd.in/g4A_wskP
Debosmit Ray (debo) replied to a comment
1w
Xu Ning and it worked out pretty nicely ha
Debosmit Ray (debo) commented on a post
2w
for teams that run software, are you seeing them being aware of the ultimate geography/hardware?
does software behave differently based on where it’s getting deployed, or is some overall level of x-cluster consistency managed by platform teams?
Debosmit Ray (debo) commented on a post
2w
yea - Steve Yegge talks about this area in this blog https://steve-yegge.medium.com/steveys-birthday-blog-34f437139cb5
(relevant area - grep for “part 2: time”)
-
Debosmit Ray (debo) reacted on thisDebosmit Ray (debo) reacted on this🌟🌟 Metaflow reached 10,000 stars on GitHub! 🌟🌟 Surely cool AI projects get to 10k stars in a week these days, but if you believe in the Lindy Law, you can be confident that Metaflow is going to be around for longer than a week too 🤗 (btw, the visualization below shows the actual distribution of stargazers over time. Calling out folks in Africa to join the community in particular and help balance the distribution! 🌍 )
-
Debosmit Ray (debo) liked thisDebosmit Ray (debo) liked thisWorking at Hebbia has been equal parts holding on for dear life and equal parts having a blast. Our customer base is growing rapidly, and we're scaling my org to (hopefully) keep up. With that, 4 new roles that we're hiring for (in addition to all the SWEs you can imagine)! Forward Deployed Engineers: https://lnkd.in/g5ignAbd Eng Manager: https://lnkd.in/gwJ_weNb Data Scientist, Product: https://lnkd.in/gYge4zb8 Data Engineer: https://lnkd.in/gwhNP67g Fine print: we're not working with agencies/head hunters, and applying directly is your best bet because I'm not keeping up with LinkedIn messages.
-
Debosmit Ray (debo) liked thisCloud Native Seattle is back for the second installment. We're talking AI this time. Come join us!Debosmit Ray (debo) liked this🚀 Announcing our second #meetup - #CloudNative in the Age of #AI Our first meetup in February was a blast (shoutout to Sameera Jayasoma and Craig McLuckie for kicking us off!), and we're back with another great lineup: Talks: 🎤 Alexander Lawrence (Technical Marketing Director @ LaunchDarkly) "Control Freaks Ship Better" 🎤 Kaslin Fields (GKE DevRel Engineering @ Google, CNCF Ambassador) "What's new in Kubernetes for running AI shenanigans?" 📅 April 16th | 5:00 PM 📍 Bellevue City Center Building (first floor conference room) 🍕 Pizza & Networking 🚗 Validated parking available RSVP → tinyurl.com/cncf-sea-apr26 #CNCF #Seattle
-
Debosmit Ray (debo) liked thisDebosmit Ray (debo) liked this🚨 CRITICAL: Active supply chain attack on axios -- one of npm's most depended-on packages. The latest axios@1.14.1 and axios@0.30.4 now pull in plain-crypto-js@4.2.1 -- a brand-new package that didn't exist before today. This is textbook supply chain installer malware. Every npm install pulling the latest version is potentially compromised right now. axios has 100M+ weekly downloads. Socket AI analysis confirms this is malware. plain-crypto-js is an obfuscated dropper/loader that: • Deobfuscates embedded payloads and operational strings at runtime • Dynamically loads fs, os, and execSync to evade static analysis • Executes decoded shell commands • Stages and copies payload files into OS temp and Windows ProgramData directories • Deletes and renames artifacts post-execution to destroy forensic evidence If you use axios, pin your version immediately and audit your lockfiles. Do not upgrade. Developing story...
-
Debosmit Ray (debo) liked thisDebosmit Ray (debo) liked this🚨🚨 axios has been compromised on npm. ⚠️ axios is one of the most popular npm packages with over 83 million weekly downloads. Multiple malicious versions (including 1.14.1 and 0.30.4) have been published. The poisoned releases bypass the project's normal CI/CD pipeline entirely. 🔍 StepSecurity AI Package Analyst detected this compromise. The malicious versions inject a hidden dependency that drops a cross platform remote access trojan targeting macOS, Windows, and Linux. The malware contacts a live C2 server and delivers platform specific payloads, then deletes itself to evade detection. 🛡️ If you have installed any of the affected versions, assume full system compromise. Pin to known safe versions and rotate all secrets immediately. 🔗 We are actively investigating and will update the blog post in the comment. #SupplyChainSecurity #npm #axios #Malware #CyberSecurity #DevSecOps #StepSecurity
-
Debosmit Ray (debo) liked thisDebosmit Ray (debo) liked thisSix months into my path as CTO at Seqera, and I’m feeling even more excited than I was on day one. Coming from a non-biotech background, I’ve been genuinely blown away by the pace of innovation and discovery happening in this space. Every day I’m learning something new — not just about the science, but about the incredible impact that the right tools can have in accelerating it. What inspires me most is seeing how Seqera and Nextflow are helping teams push boundaries, scale their work, and unlock insights that simply wouldn’t have been possible before. Grateful to be part of a team that sits right at the intersection of technology and life-changing discovery — and even more excited for what’s ahead.
-
Debosmit Ray (debo) liked thisDebosmit Ray (debo) liked thisAt the #ISMGStudio, Michael Novinson spoke with Hudson Thrift, CISO at Amazon.com to discuss; 🔹Why initiatives that can't scale beyond human use rarely justify significant investment; 🔹How autonomous penetration testing enables continuous security validation instead of periodic, point-in-time assessments; 🔹How Amazon uses red team and blue team agents to generate new threat detections faster than traditional methods. He says, "As a business, you got to work back from your customers. And if the way you're using AI isn't ultimately what they want, or isn't serving their need, then it's probably the wrong thing that you're doing." Watch the full interview- https://lnkd.in/eNSp-w_k #ISMGStudio #RSAC #Cybersecurity #ISMGNews #Pentesting #ApplicationSecurity #AppSec
-
Debosmit Ray (debo) liked thisDebosmit Ray (debo) liked thisWe 2x’d our AI code output, but nearly burned out our review cycle in the process. At Fi Money, AI-driven development this past year made our metrics look great: PR volume up. Cycle time down. But one metric told a different story: token usage. Digging deeper, a pattern emerged: Some engineers were precise. They used AI like a scalpel, tight prompts, clear context, minimal waste. Many weren’t. They brute-forced it, dumping massive context, shipping 400+ line PRs they couldn’t fully explain. A small minority remained indifferent. We were seeing two modes emerge. What Andrej Karpathy coined as vibe coding: prompt, accept, ship, was creeping in. Luckily for us our review process was catching this before it went to Production. We needed a solution and we didn’t add more seats or increase limits. We slowed things down. We introduced a simple ritual: Show & Tell. Every engineer had to walk through: The exact prompts they used The context they provided Why the output actually made sense That shift changed everything. We moved from vibe coding to agentic engineering. AI writes the code. Engineers own the reasoning. We didn’t fix this with better AI. We fixed it with better discipline. The impact was immediate: Token usage dropped Prompt quality improved Code comprehension went up Moving fast with AI is easy. Moving with precision is the real challenge. What’s the most frustrating bottleneck you’ve hit since rolling out AI coding tools? I’d love to hear how you made it work.
Experience & Education
-
DevZero
View Debosmit’s full experience
See their title, tenure and more.
or
Projects
-
University of Washington Business Plan Competition
Comprised of 4 rounds, Buerk Centers’ marquee event awards seed-funding to winning startups. Consulted for team creating technology that enables geo-tagging, facial recognition and content based search for image files.
• Created business plan with focus on monetization strategy that would enable startup to achieve exit goals
• Conducted competitive analysis of market and proposed 2 features that would help product compete effectively
• Currently competing in first round of challenge
Other creators -
-
Towers of Hanoi
Designing and implementing the Towers of Hanoi game, on an Altera DE1 board using a Cyclone II microprocessor. The implementation supported the use of 3 rings, and the player could move a chosen ring left, right or put it back down. The player would be timed from the second the first ring is touched and a message would be shown on successful completion of the game.
-
University of Washington Business Plan Competition
-
Comprised of 4 rounds, Buerk Centers’ marquee event awards seed-funding to winning startups. Consulted for team creating technology that enables geo-tagging, facial recognition and content based search for image files.
• Created business plan with focus on monetization strategy that would enable startup to achieve exit goals
• Conducted competitive analysis of market and proposed 2 features that would help product compete effectively
• Currently competing in final round of challengeOther creators -
Honors & Awards
-
Annual Dean's List
University of Washington
Other similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content