Sherv Nariman

Building from zero. One solo founder. A small audience. A big target: $15K MRR in 12 months. I’m using AI to turn real problems into focused SaaS products, then letting the market decide what survives. I’ll share the process, the numbers, and the misses.

@kubernetesio Headlamp as the path forward makes sense if the old dashboard was starting to feel like a maintenance trap

Kubernetes Dashboard to Headlamp: A Step-by-Step Guide - kubernetes.io/blog/2026/07/1… #Kubernetes

@C4708911Charles Nice release and the one-line API is exactly the sort of thing that makes map styling less painful

🚀 GeoPlot-Themes v0.1.4 is now on PyPI! 🌍 Create stunning, publication-ready maps in Python with a single line of code. Built for GeoPandas users. We'd love your feedback while it's in testing! 👇 pypi.org/project/geoplo… #Python #GIS #GeoPandas #OpenSource

@Source__Design Turning live CSS into a structured brief is a smart bottleneck remover for design handoff

Manual CSS inspection is a bottleneck for frontend workflows. With Source Design’s new CLI and MCP, you can extract design tokens directly into your terminal or through your AI agent, turning any website's live CSS into a structured DESIGN.md in seconds.

@smthomas3 @mastra The schedules bit is the one that stands out because autonomy only gets interesting once context arrives on its own

last week's @mastra recap first class skills → add skills directly to a code or file registered agent schedules → turn up the autonomy by waking up your agents and workflows with context custom signal providers → enables agents to react to events from GitHub, Slack, databases, APIs, webhooks, and more. even more to 🚢 this week...

For sure, the thing I see all the time is tests trying to be “smart” about auth instead of just running the damn flow. Stuff like stubbing current user, faking sessions, skipping controllers feels clever in the moment, but it stops testing what the app actually does. A few months later everything’s drifting and nobody knows why the suite still passes. Keeping it simple has saved me so many headaches.

@ShervNariman Do you mind sharing an example of optimizing for cleverness? I am teaching automated tests right now and would love to see examples of what not to do as well.

Rails' built-in authentication is simple by design. And it should be the same for testing it! Here's how I approach the testing for built-in auth with Minitest. 🧵 1/8

@swhan0329 The demand for real Codex workflows tracks with what I keep hearing too people want the boring process not just the magic

Demand for real-world Codex workflows is huge. Our Show & Tell drew ~4× more applications than capacity - 8 speakers sharing how they use Codex across work, AI agents, web performance, and app launches. I’ll publish 1–2 workflow stories a week starting next week. #OpenAIBuildWeek

@GithubProjects Curious how many folks actually update DESIGN.md after the first successful agent pass

The first survey on DESIGN.md, the reusable design brief for AI coding agents, based on 65K+ responses from Developers and Vibe Coders.

@hardfist_1 Yeah CSS chunking is where correctness quietly fights performance and neither side is wrong

Getting JS chunking right is already incredibly hard. Getting CSS chunking right is even harder—another classic case of the correctness vs. performance trade-off: gitnation.com/contents/turbo… BTW, Tobias Koppers’ talks are some of the best resources out there for learning about bundling: gitnation.com/person/tobias_…

@coolprobn I do. The moment auth tests start optimizing for cleverness instead of clarity, they drift from the contract they’re supposed to protect. Keeping them boring is my go to lol

@ShervNariman Do you approach your tests for authentication the same way?

@jmdenisme @ProtonPrivacy Love this path of just building the missing bits instead of living with a half finished mail workflow

I switched to the @ProtonPrivacy apps suite but was missing a few features, so I built a Brave/Chrome extension to fix the experience. Proton Mail - Auto-select next message after archive/delete - Reading size slider - Compact list - Real sender address (anti-phishing) - Hide "[N]" conversation count Proton Calendar - Show ALL events in month view (no more "N more") - ISO 8601 dates Free & open source: github.com/jmdlab/proton-…

@kr812345 This is the part people skip until production starts paging them and the magic stops feeling magical

Building AI has made me appreciate boring engineering. Everyone talks about prompts. Almost nobody talks about: Error handling Observability Fallbacks Evaluation The magic gets users. The engineering keeps them. That's where great AI products are built.

@ThePracticalDev @bendhalpern This is the part people skip until the rewrite. Why rarely survives the first refactor

Code tells you how a system works, but never why. Thoughts on the intent gap, slop describing slop, and why the machines still need your markdown files. { author: @bendhalpern } dev.to/ben/the-myth-o…

@mattpocockuk I keep relearning this the hard way too. Clever first drafts still need boring verification

Sometimes I experiment with going deep in the dumb zone to implement a feature Today it cost me 90 minutes of debugging to fix its stupid mistakes in a complex build Need to stop trying this

@Docker Yeah that combo is why I get nervous about agents that can both read untrusted stuff and take action without a hard runtime boundary

An agent doesn't have to be malicious to be dangerous. Read access + untrusted content + the ability to act = the "lethal trifecta," and it's baked into most useful agents by design. Our take from AI Engineer World's Fair on why the fix lives at the runtime → bit.ly/4fdxKFT

@PlayGameGen No purple gradients might be the most useful constraint in modern frontend

Hallmark is a design skill for Claude Code, Cursor and Codex that refuses to look AI-generated. No purple gradients. No default font stack. 5.7k stars, mostly CSS.

@GeekyJunk Love that goal. Curious which loader edge case made the architecture click for you

📷 One of the goals with 𝗦𝘁𝗶𝘁𝗰𝗵𝗿 (a small JavaScript/TypeScript bundler I'm building from scratch) is to understand why bundlers are designed the way they are, not just how to use them. This week I added a 𝗹𝗼𝗮𝗱𝗲𝗿 𝘀𝘆𝘀𝘁𝗲𝗺. 📷

@mattpocockuk I've hit that same gap. The map only helps once the words already mean the same thing to everyone

Debating a new /wayfinder change: Lots of folks are reporting that /wayfinder doesn't do as much 'shared language' work as /grill-with-docs stuff. I am also seeing this. So, I propose adding an explicit step BEFORE we make the map to firm up the language. I.e. in the very first /wayfinder mapping session, you immediately go into /domain-modeling (if needed) to establish the new entities. That way, the new language bleeds into all the rest of the tickets before they're created, making them more concise. WDYT?

@devodii_ @Chapa_ET Solid ship. How painful was it keeping the abstraction honest once Chapa’s edge cases showed up

shipped a @Chapa_ET adapter for paykit-sdk today 🎉 paykit lets you accept payments across any provider with one consistent api. now chapa's ethiopian rails are in, fully tested. africa, latam, india, and southeast asia all covered in one sdk now. open source 👇 → usepaykit.dev/providers/chapa

@gauravtoshniwal That gap feels right. Spotting weirdness is easy, trusting an AI diagnosis is a different bar

Elastic surveyed 500 teams on AI in observability. 85% use it, only 49% use it for root cause analysis. Detection and investigation are on different adoption curves. #SRE #Observability #GenAI #AIOps #IncidentResponse