Kamil's Takeaways from Cursor Cloud Agents Demo

Tightening the Dev Loop at Boosted with Cursor Cloud Agents This week, I demoed Cursor Cloud Agents at Boosted’s WTF (Weekly Tech Forum)—our recurring space to share tools, experiments, and lessons learned. The closest I’ve seen to “reviewing features, not just code” Cursor Cloud Agents let you spin up a cloud environment for a repo, prompt an agent to make a change, and then review the results: • Runs in the cloud (usable from any device) • Can do UI + test validation and produce evidence (e.g., browser runs / videos) • Opens a PR by default, guided by your rules and conventions Reducing workflow friction This is especially useful for the “small but annoying” stuff that steals time: • Quick fixes while you’re away from your dev setup • Small refactors, config updates, and cleanup • Verifying UI flows without manually clicking everything The outcome: faster iteration, fewer context switches, and less time spent on repeatable work—while keeping quality gates (PR review, tests) intact. If you’re experimenting with agent workflows or skills, I’d love to compare notes!

I've been using this MCP for months—it's one of the small tools that actually ships. Your agent generates images inside Cursor. No switching tabs. No copy-paste. Flux models, open source, saves locally. Use cases for me: infographics, blog posts, gamedev assets, mockups, prototyping. Anything where I need an image that matches the idea. Setup: Replicate API token, add to global MCP config, done. Works with Cursor, Cloud Code, Codex. Cost tracking built in—about 3¢ per image with Flux 2 Pro. Sharing because it's underrated. Flux Replicate MCP on NPM if you want to try it.

here is the pattern i keep seeing. a team picks a paas. the pitch is clean: no infrastructure to manage, just push and deploy. they get started. things work. they stop thinking about the platform. six months later they are debugging timeout issues on background jobs. or hitting undocumented connection limits that only appear at a certain request volume. or waiting 48 hours for a support response from a chatbot that cannot answer infrastructure questions. the managed part was never fully true. it was true for the median team with the median use case. any deviation from that and you start pulling back the curtain. the tool isn't the problem. the category assumption is: that managed means no infrastructure knowledge required, no compliance risk, no surprises. it never did. it just deferred the complexity until a more inconvenient moment. you're solving the same risk with the same architecture every time you migrate between managed platforms. the ceiling moves. the dependency stays. we are building something that actually earns the claim. a small group of engineers is already inside. https://azin.run

most teams don't decide to adopt kubernetes. they inherit it. someone on the team used it at their last job. or the cloud provider recommended it. or a blog post made it sound inevitable. and then suddenly you have 12 to 18 yaml files before your first service goes live. each file made sense when someone added it. deployments.yaml, services.yaml, ingress.yaml, configmap.yaml, horizontalpodautoscaler.yaml. each one solving a real problem. together they form a maintenance surface area nobody planned for and nobody has time to fully understand. then comes the cluster upgrade. the cert renewal. the mysterious pod that keeps crashing in staging but not in prod. the on-call rotation for the thing you just wanted to run your app on. the tool was built for companies with dedicated platform teams and years of operational investment. most startups using it have neither. that is not a people problem. the people are smart. it is a category error. and it is costing teams more than they can measure in hours, not cloud credits. we are working on what it looks like when you get all the control without carrying all of that. you can try it out now. https://azin.run

You've got to slice your MCPs. Today I heard Cyclr compared to a bread slicer, and I loved it - and I have unashamedly stolen the concept. An MCP issue being discussed in the MCP echo chamber is Context bloat (providing too much info to the LLM and burning too many tokens). It is becoming a real consideration for advanced MCP users. A Day Two challenge of rolling-out MCPs. How you roll them out efficiently and scalably. In-built into an MCP PaaS is a 'slicer' that helps manage the bloat (don't eat all the bread!). An MCP PaaS enables you to create discrete MCPs with discrete purposes and then to create unique URLs for each and every unique client - all automatically. For example: Eating the whole loaf = enabling your entire API and all its' methods in one MCP. A huge amount of bread to consume that costs tokens and causes bloat. Slicing the loaf = creating a discrete MCP server per function or capability. For example an MCP server that solely 'Gets Contacts' whilst another MCP server 'Checks Subscription Status'. Each containing a single method. Less bread to consume, one slice at a time, less tokens, less bloat. To do this you need a framework that helps you scale your MCP and efficiently create and deliver the slices. For MCPs to work for you, and moreover to work for your customers, they need to deliver and to deliver fast and efficiently. You need a slicer. That is ONE of the functions of an MCP PaaS. Tap us up for a direct chat on the features of an MCP PaaS if you want to learn more.

Cursor cloud agents and Cursor automation have sped my code review process by at least 3X. It identifies whats possible within the scope of the PR, it is sometimes too ambitious or unrealistic with its asks but thats still manageable through some clear instructions. My next task is to get Cursor automation to fix prioritized Sentry issues based on some ranking mechanism. We aim to have some bugs ready for human reviews for sanity checks once the agents have fully validated the fixes. The only doubt I have is in the ability to replace that whole setup with some self hosted infra in the future without making a mess.

You don’t need a cloud account to understand Kubernetes. You never did. Before tools like Minikube, getting started with Kubernetes meant setting up clusters manually, dealing with multiple nodes, and often relying on cloud infrastructure just to practice. For beginners, that was a big barrier. You ended up spending more time on setup than actually learning Kubernetes. That’s where Minikube changes things. It lets you run a real Kubernetes cluster locally on your machine. Not a mock, not a simulation, but an actual cluster that behaves very close to production environments (single-node by default, but can scale to multiple nodes). Now the focus shifts. Instead of worrying about infrastructure, you can actually learn what matters: Deployments, Services, Ingress, ConfigMaps, Secrets, and Persistent Volumes. All running locally, all testable in real time. What makes Minikube interesting is not just learning. It’s practical. You can test your deployments before pushing to the cloud, debug configs safely, and experiment freely without worrying about cost or breaking anything in production. It also helps you understand how your application will behave in a Kubernetes environment before going live. And this changes the mindset completely. Kubernetes stops feeling like a “cloud-only” tool. It becomes something you can run, break, fix, and actually understand on your own machine. That’s powerful. You can take a full app, containerize it, deploy it locally, and see how everything connects. Then later, move the same setup to a real cluster with much more confidence. There are other local tools like Kind and k3d, but Minikube is still one of the most beginner-friendly ways to get started. If you’re learning Kubernetes, Minikube removes the biggest barrier. #Kubernetes #Minikube #DevOps #CloudNative #SoftwareEngineering

A client asked me how to get their Nano Banana API Key and I figured you might have this question too so I made carousel SOP for it. Google changed how you get API keys recently it used to be a mess. You had to go into the Google Cloud Console, manually create a project, hunt down the “APIs & Services” section, enable the Generative Language API by name, then navigate to the Credentials page to actually generate a key. Then you had to setup billing! Half the time you’d get lost in menus that were built for enterprise DevOps teams, not builders who just want to plug something into their app and move on and now it’s all in one place. Here it is 👇

In the first of a four-part series, Shir Meir Lador outlines how to build Dev Signal, a multi-agent system designed to automate technical content creation by identifying trending questions on Reddit and grounding them in official documentation. The core of the project focuses on using the Model Context Protocol (MCP) to standardize how AI agents interact with external tools and data: - Discovery: A Reddit MCP server allows the agent to find high-engagement technical questions. - Grounding: The Developer Knowledge MCP server enables the agent to perform semantic searches across official Google Cloud documentation for accuracy. - Multimodal Generation: A custom "Nano Banana" MCP server uses Gemini 3 Pro Image to generate professional infographic headers for the blog posts. Read it here: https://lnkd.in/duWHgSjV

𝗔𝗪𝗦 𝗕𝗶𝗹𝗹𝗶𝗻𝗴 𝗗𝗮𝘀𝗵𝗯𝗼𝗮𝗿𝗱 𝗠𝗼𝘀𝘁 𝗔𝗪𝗦 𝗯𝗲𝗴𝗶𝗻𝗻𝗲𝗿𝘀 𝗼𝗽𝗲𝗻 𝘁𝗵𝗲 𝗰𝗼𝗻𝘀𝗼𝗹𝗲... 𝗮𝗻𝗱 𝗽𝗮𝗻𝗶𝗰 𝘁𝗵𝗲 𝗺𝗼𝗺𝗲𝗻𝘁 𝘁𝗵𝗲𝘆 𝘀𝗲𝗲 𝘁𝗵𝗲 𝗕𝗶𝗹𝗹𝗶𝗻𝗴 𝗗𝗮𝘀𝗵𝗯𝗼𝗮𝗿𝗱. So many numbers. So many services. So many graphs. 𝗕𝘂𝘁 𝗵𝗲𝗿𝗲’𝘀 𝘁𝗵𝗲 𝗿𝗲𝗮𝗹𝗶𝘁𝘆 👇 The 𝗔𝗪𝗦 𝗕𝗶𝗹𝗹𝗶𝗻𝗴 𝗗𝗮𝘀𝗵𝗯𝗼𝗮𝗿𝗱 is simply your 𝗰𝗹𝗼𝘂𝗱 𝗲𝘅𝗽𝗲𝗻𝘀𝗲 𝗰𝗼𝗻𝘁𝗿𝗼𝗹 𝗿𝗼𝗼𝗺.  • Think of it like the 𝗲𝗹𝗲𝗰𝘁𝗿𝗶𝗰𝗶𝘁𝘆 𝗺𝗲𝘁𝗲𝗿 𝗶𝗻 𝗮 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴.  • You don’t control electricity from the meter…  • but it tells you 𝘄𝗵𝗼 𝗶𝘀 𝗰𝗼𝗻𝘀𝘂𝗺𝗶𝗻𝗴 𝗽𝗼𝘄𝗲𝗿 𝗮𝗻𝗱 𝗵𝗼𝘄 𝗺𝘂𝗰𝗵. 𝗔𝗪𝗦 𝘄𝗼𝗿𝗸𝘀 𝘁𝗵𝗲 𝘀𝗮𝗺𝗲 𝘄𝗮𝘆. Every EC2 instance, S3 bucket, or database you run sends its 𝗰𝗼𝘀𝘁 𝘀𝗶𝗴𝗻𝗮𝗹𝘀 to the billing dashboard. 📌 𝗪𝗵𝗮𝘁 𝘁𝗵𝗲 𝗕𝗶𝗹𝗹𝗶𝗻𝗴 𝗗𝗮𝘀𝗵𝗯𝗼𝗮𝗿𝗱 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝘀𝗵𝗼𝘄𝘀 ✅ 𝗖𝘂𝗿𝗿𝗲𝗻𝘁 𝗠𝗼𝗻𝘁𝗵 𝗖𝗼𝘀𝘁 Real-time estimate of what you’re spending right now. ✅ 𝗦𝗲𝗿𝘃𝗶𝗰𝗲-𝘄𝗶𝘀𝗲 𝗕𝗿𝗲𝗮𝗸𝗱𝗼𝘄𝗻 See which services (EC2, S3, RDS) are generating cost. ✅ 𝗙𝗼𝗿𝗲𝗰𝗮𝘀𝘁𝗲𝗱 𝗦𝗽𝗲𝗻𝗱 AWS predicts your end-of-month bill. ✅ 𝗖𝗼𝘀𝘁 𝗘𝘅𝗽𝗹𝗼𝗿𝗲𝗿 𝗔𝗰𝗰𝗲𝘀𝘀 Deep analysis of usage trends. ✅ 𝗕𝗶𝗹𝗹𝗶𝗻𝗴 𝗔𝗹𝗲𝗿𝘁𝘀 Set notifications if spending crosses limits. ⚙️ 𝗪𝗵𝘆 𝘁𝗵𝗶𝘀 𝗺𝗮𝘁𝘁𝗲𝗿𝘀 𝗳𝗼𝗿 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀  1. Cloud mistakes are expensive.  2. Launching one large EC2 instance accidentally.  3. can cost more than a full month of learning.  4. Good engineers don’t just deploy infrastructure.  5. They 𝗺𝗼𝗻𝗶𝘁𝗼𝗿 𝗰𝗼𝘀𝘁 𝗶𝗺𝗽𝗮𝗰𝘁 𝘁𝗼𝗼. 𝗨𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝘁𝗵𝗲 𝗕𝗶𝗹𝗹𝗶𝗻𝗴 𝗗𝗮𝘀𝗵𝗯𝗼𝗮𝗿𝗱 𝗶𝘀 𝗼𝗳𝘁𝗲𝗻 𝘁𝗵𝗲 𝗳𝗶𝗿𝘀𝘁 𝘀𝘁𝗲𝗽 𝘁𝗼𝘄𝗮𝗿𝗱 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴: 🚀 Cost Optimization 🚀 FinOps 🚀 Production Cloud Management 𝗔𝗻𝗱 𝘁𝗵𝗮𝘁’𝘀 𝘄𝗵𝗮𝘁 𝘀𝗲𝗽𝗮𝗿𝗮𝘁𝗲𝘀 𝗰𝗼𝗻𝘀𝗼𝗹𝗲 𝘂𝘀𝗲𝗿𝘀 𝗳𝗿𝗼𝗺 𝗰𝗹𝗼𝘂𝗱 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝘀. 𝗙𝗼𝗹𝗹𝗼𝘄 𝗼𝘂𝗿 𝗟𝗶𝗻𝗸𝗲𝗱𝗜𝗻 𝗣𝗮𝗴𝗲 𝗳𝗼𝗿 𝗱𝗮𝗶𝗹𝘆 𝗰𝗹𝗼𝘂𝗱 𝗰𝗹𝗮𝗿𝗶𝘁𝘆: https://lnkd.in/dN4JSkfH 𝗝𝗼𝗶𝗻 𝗼𝘂𝗿 𝗪𝗵𝗮𝘁𝘀𝗔𝗽𝗽 𝗖𝗹𝗼𝘂𝗱 𝗖𝗼𝗺𝗺𝘂𝗻𝗶𝘁𝘆: https://lnkd.in/dTJfEFyK 𝗪𝗲𝗯𝘀𝗶𝘁𝗲: www.vyomanant.com 𝗘𝗺𝗮𝗶𝗹: academy@vyomanant.com #VyomanantAcademy #Vyomanant #CloudComputing #DevOps #CloudCareer #TechCareers #AWS #AWSBilling #AWSCostManagement #FinOps #CloudLearning #CloudEngineering