Performance98 on Lighthouse, Slow for Real Users: Core Web Vitals
The developer kept sending screenshots of the score. Real users had a median LCP of 4.6 seconds, because the consent script blocked rendering for two of them.
Read MoreThe Blog
Thoughts, tutorials and notes on backend engineering & the cloud.
AllDatabases 4API Development 3AI Agents 2AI Coding Tools 2Artificial Intelligence 2DevOps 2Freelancing 2Performance 2AI Agents vs Chatbots 1AI Automation 1AI Engineering / Machine Learning 1AI Infrastructure 1AI Observability 1AI Security 1Authentication 1Backend Development 1Blockchain 1Career 1Cloud Computing 1Cloud Security 1Context Engineering 1Cybersecurity 1Data Engineering 1Data Science 1DevSecOps 1Docker 1Edge Computing 1Emerging Technologies 1Engineering Metrics 1Feature Flags 1FinOps 1Generative AI 1Git 1Interviews 1IoT 1Kubernetes 1Large Language Models 1MCP 1MVP 1Machine Learning 1Message Queues 1Microservices 1Node.js 1Observability 1Platform Engineering 1Prompt Engineering 1Quantum Computing 1RAG 1React 1Real-Time 1Robotics 1Rust 1SRE 1SaaS Architecture 1Small Language Models 1Software Development 1System Design 1Testing 1TypeScript 1Vibe Coding 1Web Development 1Web Fundamentals 1WebAssembly 1Zero Trust 1
Performance
Backend DevelopmentNode vs Django vs Go: The Argument Is Usually the Wrong One
A founder wanted me to settle it. The product was a dashboard with two hundred users, and the entire performance argument was about a bottleneck that did not exist.
Read More
SREProduction Incidents: I Made My First Serious One Worse
Twenty minutes of confident action, all of it wrong, because I diagnosed in the first ninety seconds and spent the rest defending that guess.
Read More
DevOpsFrom 24 Minutes to 6: Making CI Fast Enough to Wait For
Nobody waited for our 24-minute pipeline. They pushed, switched tasks, and came back an hour later — by which time the context was gone.
Read More
FreelancingWorking With International Clients From Pakistan
A UK client asked why they should hire someone in Lahore over Manchester. I answered with cost, which was wrong, and we did not work together.
Read More
FreelancingHow I Actually Got Freelance Clients as a Developer
My first proposal was 1,400 words about me. Forty of those got two replies, both asking for a lower price. The one that worked was 150 words about them.
Read More
CareerThe Backend Developer Roadmap I'd Actually Follow in 2026
Someone sent me a roadmap with two hundred boxes and asked how long it would take to finish. Nobody finishes that, and the order is wrong anyway.
Read More
API DevelopmentHTTP Status Codes You Are Probably Using Wrong
A partner retried a malformed request forever because we returned 500 instead of 400. Both sides behaved correctly and the system still broke.
Read More
API DevelopmentRate Limiting: Four Algorithms and 1.4 Million Requests
A customer's retry loop made 1.4 million requests overnight and took the API down. They were not attacking us — their code had a bug and we had no limits.
Read More
Testing94% Coverage and Broken Weekly: Unit vs Integration vs E2E
Every service method had a test with the repository mocked. The tests passed whether or not the SQL was valid. Coverage measured how much code ran.
Read More
System DesignScaling to Your First Million Users, One Stage at a Time
Six microservices, Kafka, sharded databases — and nine hundred users. Shipping a change to a form took three weeks.
Read More
Node.jsThe Node.js Event Loop, and Why One Endpoint Froze Everything
CPU pinned at 100% on one core, seven idle, database untouched. One endpoint parsed a CSV synchronously — 300ms during which every request was frozen.
Read More
GitGit: How to Undo Almost Anything
"I ran git reset --hard and lost a day of work." He had lost twenty minutes. Getting the rest back took about forty seconds with reflog.
Read More
Message QueuesKafka vs RabbitMQ vs SQS: Picking a Message Queue
A team ran Kafka for eighteen months to move four thousand messages a day. That is three a minute. A Postgres table would have done it.
Read More
PerformanceCaching Strategies, and the Four Places We Forgot to Invalidate
A price change went out and one customer kept seeing the old one. It was cached in four places. We had invalidated one of them.
Read More
DatabasesWhy Your SQL Query Is Slow: Six Causes, In Order
"The database is slow" is the least useful sentence in backend work. The database is almost never slow — a specific query is, for one of about six reasons.
Read More
DatabasesDatabase Indexing Explained: Fast Until Suddenly It Isn't
A page loaded in 200ms for a year, then took nine seconds. Nothing was deployed. The table had simply crossed four million rows.
Read More
Web FundamentalsCORS Explained: Your Request Was Not Actually Blocked
A developer spent a day on one CORS error. The top search results tell you to install a browser extension, which is the worst of the available fixes.
Read More
AuthenticationJWT vs Sessions: Stop Reaching for Tokens by Default
Support asked me to log out a compromised account. The answer was: we cannot. That token was valid for six more days and nothing could stop it.
Read More
InterviewsHow I Actually Answer a System Design Interview
I failed my first one by drawing boxes ninety seconds in. Fifteen minutes later the interviewer asked how many users the system had, and I had never asked.
Read More