👾Travis Johnson

If your platform team still calls itself **DevOps**, you're already behind. The best platform teams I've worked with don't just ship pipelines- they ship experiences. Their success isn't measured by uptime, it's measured by how fast developers can deploy safely without asking for help. The shift is subtle but fundamental: Traditional DevOps = gatekeepers of infra Modern Platform Teams = product owners of developer experience (DevEx) As organizations scale, DevOps evolves into Platform Engineering, a layer that provides reusable building blocks for delivery: - IaC modules, - CI/CD templates, - Oobservability dashboards, - Policy-as-code, - Secrets management, - and Standardized runtime patterns. The best teams have roadmaps, user feedback loops, and service-level objectives- just like any customer-facing product. To scale your platform team effectively: ✅  Treat your developers as customers, not requesters. ✅  Measure platform adoption, not ticket closure rates. ✅  Build feedback into your GitOps or portal layer, friction is a feature request. Is your platform team building products for developers or policing them? What would change if you rebranded DevOps internally as "Platform as a Product"?

9

Does your CI take longer than your standup? I once worked in a team whose pipeline used to take over 30 minutes to run. It wasn't just slow, it was a liability. Incident fix? Delayed. Deploy retry? Hope you were not in a hurry. Engineers were either stuck waiting or trying to outsmart a flaky pipeline and everything moved at the speed of frustration. Eventually, we cleaned it up. We brought the build time down to 7-10 minutes, with retries under 5. That single change transformed our pace. Deploys became smoother, fixes went out faster, and engineering morale? Way up. CI isn't just about builds. It's your feedback loop and confidence multiplier. Your safety net when things go down the hill. And if it drags, so does your team. Fix your CI. Or get used to shipping excuses instead of features.

11

5 Comments

MCP isn't the endgame for agent integration. It's the SOAP of the agentic era. The friction is real: - Context bloat is brutal. Every MCP server costs tokens before your agent even starts thinking. Anthropic and community open source projects trying to tackle that but still. - Security and auth are an afterthought. Enterprise-grade, it is not. - Ecosystem fragmentation. The adoption is increasing but still companies like Google have been exploring other options when it comes to agents integration/interoperability (ex: A2A). So what comes next? CLI-first, code-execution-native integrations (sometimes interfacing with APIs). Look at what Google did with the Workspace CLI. That's not a developer tool, that's a blueprint. A well-scoped CLI is more composable than any MCP server, reuses the entire existing package ecosystem, and is already understood by every agent runtime that can write and run code. The unlock isn't a new protocol. It's treating the shell as the integration layer. The catch? Sandboxing becomes the critical infrastructure piece. If your agent can execute arbitrary CLI calls, you need hard isolation, scoped permissions, and tight control over what surfaces get exposed. The security model can't be an afterthought. This is the same story we've told before. SOA solved "coupling". REST solved "discoverability". GraphQL solved "over-fetching". gRPC solved "performance". Each generation fixed the sins of the last. CLI + sandbox execution is next, shaped by the constraints of agent runtimes rather than human developers. The integration platform that wins won't just connect things. It'll define what agents are trusted to touch.

19

1 Comment

I just started using Claude Code (20x) and there were/are a lot of people FREAKING OUT about how Boris Cherny (Claude Code lead) uses CC. It's a great post, but I think maybe the value is for folks that aren't engineers because basically the spirit of the post is "do a bunch of shit in parallel and always use the most expensive shit" which ah...I wholeheartedly agree with. There're no secrets though and again the ceiling is Opus' performance and nothing else. The dev tooling is useful, but the only reason any of this is even remotely useful is because of SOTA models like Opus 4.5. So for the millionth time the whole industry should be focused on improving models and not tooling. The tooling is great, but it's just a symptom of flagship model ceiling perf.

2

1 Comment

Six months of decisions that never got made. Here's what that actually cost. The plan was ready. The team was ready. Four exec forums. No outcome. Just “let’s review next quarter.” While everyone waited:: • $300K burned on idle contractors • A critical platform fell out of support • Senior engineers disengaged And still, progress was expected. Indecision is still a decision. It just hides the cost. What changed things wasn’t alignment. It was ownership. “We’re starting Monday unless someone stops it by Friday.” That single shift forced clarity. Delay was no longer invisible. Where is delay costing more that deciding right now? Most leaders don’t see the cost until it’s too late.

53

41 Comments

There's one massive red flag that I've seen with almost every client of mine. The team, which includes senior engineers, takes on tech debt because of a tight deadline. They always, without exception, say that they will address the tech debt immediately after the deadline. It's a lie. They won't. There will be a new deadline, and new pressure, thus MORE tech debt. In the meantime, the tech debt from the first piece of work has already grown. And nobody will ever understand or accept your slowing down. Do not take on tech debt. Ever. Either negotiate the deadline or the scope. Never quality. You are trusted to know what you are doing. You will be accountable for the ever growing pile of steaming... Spaghetti.

55

14 Comments

I often ask, “Why did you design it this way?” The strongest candidates don’t jump to tools — they explain the trade-offs. Example: “We used EMR for this pipeline because Glue jobs were hitting concurrency limits, and we needed better tuning control.” That one line shows experience, ownership, and context awareness. #DataEngineering #InterviewTips

4

MongoDB Tip of the week... TTL Indexes. I know I have missed a few weeks.. It has been crazy busy here lately. Time to Live (TTL) Indexes are a great tool for your age off strategies. Just a few things to remember. 1. They only work on time stamps and ISODates. 2. They only work, if they can complete the deletes within the 60 second intervals TTL indexes run at. I have seen both of these issues recently. For number 1, I have seen this where at a first glance it looked fine. But looking further it became clear. Here is a quick run down. Actual index. {   v: 2,   key: { timestamp: 1 },   name: 'timestamp_1',   background: true,   expireAfterSeconds: 604800  } field in documents: { ....   timestamp: Long('1596948240000'), .... } In this example a quick look you might actually miss the fact that the field is NOT a timestamp or an ICODate. It is a long, thus the TTL index was never cleanup any of this data. It only came to my attention once the collection's size got to a size that cause me to look further at the collection. Note, this collection and its indexes predate me coming on board. This is a great example when a schema could have help to avoid this issue. Again I can NOT stress enough the importance of creating a schema. I also advise against using field names of data types, reason being they can make one make assumptions that are not true. In this example the field name, probably hide the fact that the field was NOT a timestamp. For the number 2. We have a collection that is really large.. And grows at an extremely fast pace. The TTL index can NOT perform the cleanup in the 60 seconds needed. In these cases a custom built piece of code can be written to do the heavy lifting. In these examples they might work at the beginning, but as the collection grows and the load increases, you might have to re-think your strategy. So be diligent and look at your load rates, and collection sizes. Check to make sure your TTL indexes are still working.

1

I now have my own operating system (like Windows 11 or iOS) that integrates into my own cloud. Here is an update on Vibe Coding with Claude. Overall, it's not looking great for most developers. I've moved from my cloud environment (email, doc storage, credentialing, etc) to the operating system level, an area I have little expertise. The current result is a functioning OS that is integrated into my cloud architecture for authentication that I can develop from my phone, commit the code from anywhere and have a new ISO file waiting in S3 in about 10 min. There were limitations, though. An early architectural decision proved wrong and required a full restart of the OS project. This suggests that while the model is capable of assisting with implementation, architectural guidance still needs human oversight. There were also a few instances where the model appeared overly confident while pursuing an incorrect path. In those cases, the changes had to be manually reversed. More broadly, domain knowledge remains essential. The model accelerates implementation, but understanding the underlying technologies is still necessary in order to guide it effectively. I've had to pick up quite a bit about OS drivers and greeter's this weekend. Token generation speed did not meaningfully constrain progress. The dominant time cost remains verification: ensuring that systems actually work as intended. Claims that higher token throughput materially improves outcomes are greatly overexaggerated. Finally, small implementation issues still arise and require debugging and manual correction. All in all, the technology is powerful, but does not eliminate the need for technical judgment quite yet. Next step, connecting local SLMs and cloud based LLMs into the operating system so the whole thing is AI driven. #AI #VibeCoding #Tamaton #LLM #ClaudeCode #Opus #Anthropic

38

14 Comments

via Andre Lockhart Here is the 3-step software engineering technical vetting process for hiring engineers who write great code, with a quickness. There are clearly measurable assessment criteria for each step. Use the same tech stack they'll be working in. 1. Engineer 1: Give them a sheet of paper or a screen filled with messy code and have them talk through the refactor. 2. Engineer 2: Pair with the candidate to find/fix a bug. 3. Engineer 3: Pair with the candidate to add a feature.

1

Kubernetes just announced something that affects production systems everywhere. NGINX Ingress Controller is being retired in March 2026. No more updates, no security patches, nothing. This isn’t about the project failing. It’s about maintainer burnout. For years, one or two people maintained one of the most deployed components in Kubernetes, working nights and weekends. They couldn’t sustain it. Here’s what this really means: the project that powers billions of requests across hospitals, banks, and SaaS platforms worldwide just got a death sentence because the open source contribution model broke down. The replacement is Gateway API. It’s been generally available since October 2023 and does everything NGINX Ingress did, but better. More flexible routing, better multi-team collaboration, standardized configuration that works across implementations. You get the same monitoring capabilities you had with NGINX Ingress. Prometheus metrics for latency, error rates, traffic patterns. Access logs. Pre-built Grafana dashboards. The observability story is actually stronger because Gateway API implementations like Envoy Gateway, Istio, and Traefik follow more consistent patterns. The real shift is architectural. NGINX Ingress used vendor-specific annotations for advanced features. Gateway API makes those features standard and portable. You’re not locked into one implementation anymore. But there’s a catch: you have until March 2026 to migrate, and if you’re running critical infrastructure, you need to start planning now.

13

2 Comments

MSK (Managed Kafka) looks simple on the bill until you pull it apart. Most teams know they pay for broker hours, which range from ~$0.043/hr for a kafka.t3.small to over $10/hr for a kafka.m5.24xlarge. A common production setup with three kafka.m5.large brokers costs around $460/month — that's 3 brokers × $0.21/hr × 730 hours — just for compute. However, many overlook the ongoing costs of storage at $0.10 per GB per month and the potential charges for data transfer, which is $0.01 per GB each way for cross-Availability Zone (AZ) traffic. An often missed aspect is the waste incurred from idle clusters. If no messages are processed over a seven-day period, as detected by CloudWatch, the cluster is effectively idle — and unlike EC2, MSK clusters cannot be stopped, only deleted. Common scenarios that lead to idle clusters include deprecated streaming pipelines or development clusters mirroring production setups left running without a purpose. Taking a moment to regularly review cluster activity can prevent unnecessary charges and ensure that resources align with current application needs. CloudWise detects idle MSK clusters automatically so you don't have to audit them manually. #AWS #AWSMKS #CloudWise #CostOptimization #FinOps

5

1 Comment

Totally agree, this is exactly it: AI (and humans) don’t need more signals, they need decision-grade information. Without the who, what, where, when, and the operating context, models become guessing machines. The tank story nails it: it wasn’t “high”, it was “how much time do we have, with what confidence”.

8

"We've escalated it to dev." If you've ever cringed hearing your team say this to a customer, you're not alone. Your customer called because their risk reports aren't generating. They need the data before markets open. And your first response is to tell them you're passing their problem to someone else? Here's what the customer actually hears: "This isn't my problem anymore. Good luck." During my time leading global support at a major fintech company, we supported 900+ enterprise clients running complex risk calculations. When reports failed or data looked wrong, hearing "escalated to dev" became our customers' least favorite phrase. Because escalation became abdication. 𝗧𝗵𝗲 𝗿𝗲𝗮𝗹 𝗰𝗼𝘀𝘁 𝗼𝗳 "𝗲𝘀𝗰𝗮𝗹𝗮𝘁𝗲𝗱 𝘁𝗼 𝗱𝗲𝘃": • 🔄 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗹𝗼𝘀𝘀 𝘁𝗮𝘅: Each handoff loses crucial details. By the third engineer, nobody remembers the customer mentioned this started after yesterday's market close. • ⏱️ 𝗘𝗳𝗳𝗼𝗿𝘁 𝗺𝘂𝗹𝘁𝗶𝗽𝗹𝗶𝗰𝗮𝘁𝗶𝗼𝗻: What should be a 30-minute fix becomes a 3-day investigation because teams keep starting over. • 🔍 𝗢𝘄𝗻𝗲𝗿𝘀𝗵𝗶𝗽 𝘃𝗮𝗰𝘂𝘂𝗺: "It's not our team" becomes the default response instead of "let me figure this out." • 💸 𝗧𝗿𝘂𝘀𝘁 𝗲𝗿𝗼𝘀𝗶𝗼𝗻: Every "escalation" makes customers question whether anyone actually owns their problem. 𝗪𝗵𝗮𝘁 𝘄𝗲 𝗰𝗵𝗮𝗻𝗴𝗲𝗱: Instead of escalating cases, we created 𝗰𝗼𝗹𝗹𝗮𝗯𝗼𝗿𝗮𝘁𝗶𝗼𝗻 𝗽𝗮𝗿𝘁𝗻𝗲𝗿𝘀𝗵𝗶𝗽𝘀. When support couldn't resolve a report issue, they didn't transfer ownership—they brought in a quant developer as a collaborator while staying on the case. The customer always had one point of contact. Behind the scenes, we had the right technical expertise. Result: 78% reduction in multi-team escalations and our first-contact resolution rate for technical issues jumped from 31% to 76%. 𝗦𝘁𝗼𝗽 𝘁𝗲𝗹𝗹𝗶𝗻𝗴 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿𝘀 𝘆𝗼𝘂'𝗿𝗲 𝗲𝘀𝗰𝗮𝗹𝗮𝘁𝗶𝗻𝗴 𝘁𝗵𝗲𝗶𝗿 𝗽𝗿𝗼𝗯𝗹𝗲𝗺. Start telling them you're bringing in additional expertise to solve it faster. Your customers don't want to hear about your org chart. They want to hear about their solution. This is exactly the kind of operational breakdown that kills customer trust—and why I developed 𝘁𝗵𝗲 𝗦𝘁𝗿𝗮𝘁𝗲𝗴𝗶𝗰 𝗔𝗜-𝗣𝗼𝘄𝗲𝗿𝗲𝗱 𝗢𝗽𝗲𝗿𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗘𝘅𝗰𝗲𝗹𝗹𝗲𝗻𝗰𝗲 𝗳𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸 (https://lnkd.in/e8DZCQKy). When your handoffs are broken, AI won't fix them. You need to redesign the process first. 𝘞𝘩𝘢𝘵 𝘱𝘩𝘳𝘢𝘴𝘦𝘴 𝘥𝘰 𝘺𝘰𝘶𝘳 𝘵𝘦𝘢𝘮𝘴 𝘶𝘴𝘦 𝘵𝘩𝘢𝘵 𝘮𝘢𝘬𝘦 𝘺𝘰𝘶 𝘤𝘳𝘪𝘯𝘨𝘦? 𝘚𝘩𝘢𝘳𝘦 𝘺𝘰𝘶𝘳 "𝘯𝘦𝘷𝘦𝘳 𝘴𝘢𝘺 𝘵𝘩𝘪𝘴 𝘵𝘰 𝘢 𝘤𝘶𝘴𝘵𝘰𝘮𝘦𝘳" 𝘮𝘰𝘮𝘦𝘯𝘵𝘴 𝘣𝘦𝘭𝘰𝘸. #CustomerExperience #SupportOperations #FinTech #SaaS #CustomerSuccess

16

1 Comment

The ROI isn’t just speed—it’s transfer. Prompting patterns I used for: - Survey design → also work for interview guides. - Terraform reviews → also work for policy checks. - Meeting digests → also work for PRDs. Learn once, apply everywhere.

5

I think it’s a fair observation. I started with the thinking that AI is a force multiplier. Then I learnt it’s not a scalar and rather a vector entity. Meaning it can take you dramatically far in any direction, positive (impactful) or negative (wasteful).

13

I built another thing, again a utility to help review software. I've worked on at least 5 similar tools that showed a code reviewer the terraform plan. This one is my favorite - a few lines of yaml and you get a nicely formatted PR comment with the plan. Subsequent pushes to that branch will update the comment with the new plan. Check out the project or try it out: https://lnkd.in/g-ceADiV

61

2 Comments