Engineering MetricsDORA Metrics in the AI Era: Bigger PRs, Slower Delivery
A director asked me for lines of code per developer. Our PRs got 40% larger with AI tools and delivery barely moved — which is the whole point.
Read MoreThe Blog
Page 2 of 4 — notes on engineering, AI and 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
Engineering Metrics
Real-TimeWebSockets, SSE or Polling: Picking the Right Real-Time Transport
A dashboard polled six endpoints every three seconds — 24,000 requests a minute, 99% of them identical. The fix was not WebSockets for all six.
Read More
SaaS ArchitectureMulti-Tenant SaaS: The Worst Sentence in a Bug Report
"A user saw another company's data." One reporting endpoint filtered by date and status but not by tenant. Every other query in the codebase was correct.
Read More
AI Engineering / Machine LearningLangChain vs Direct API Calls: When to Use Each
Should you use LangChain or just call the API? The answer shapes your project's complexity, performance and portability. Here is how I actually decide.
Read More
MVPHow to Build an MVP with AI in 2 Weeks
Two weeks used to buy a clickable prototype and an argument about the schema. With AI in the loop it can buy a working MVP — if the process is right.
Read More
AI Agents vs ChatbotsAI Agents vs Chatbots: What's the Real Difference?
Everyone calls everything an AI agent now. There is a real, useful distinction between a chatbot and an agent, and it changes what you should build.
Read More
Vibe CodingVibe Coding
"Vibe coding" is half a joke and half a real shift in how software gets built. What actually changes — and why deep technical skill matters more, not less.
Read More
Artificial IntelligenceThe Future of AI in Business
For years "AI in business" meant chatbots and dashboards. That phase is ending — AI is starting to own whole processes, not just the tasks inside them.
Read More
Feature FlagsFeature Flags: One Lever Is Not Enough
A Friday pricing change went to every customer at once. Eleven support tickets in twenty minutes, and rollback took forty because a migration shipped with it.
Read More
ObservabilityOpenTelemetry: From Two Days of Grepping to Four Minutes
A checkout took eleven seconds one request in twenty, across four services, all of them green. Grepping four log streams by timestamp took two days.
Read More
AI SecurityAI Red Teaming: I Broke My Own Feature in Four Minutes
I filed a support ticket containing text that read like instructions. The summariser followed them and listed other customers' tickets, in a tidy bulleted format.
Read More
DatabasesJust Use Postgres: Six Data Systems for 4GB of Data
Postgres, MongoDB, Redis, Elasticsearch, RabbitMQ and Pinecone — for a six-person team with 4GB of data. Nobody made a bad decision individually.
Read More
AI InfrastructureServerless GPUs: Paying $1,100 a Month for 40 Minutes of Work
Their GPU sat idle 23 hours a day because cold starts took three minutes. That trade — pay for idle or wait for load — is the whole economics of inference.
Read More
WebAssemblyWebAssembly in Production: When It Is Actually the Right Tool
Users uploaded 400MB videos so a server could shrink them to 30MB. Moving the work into the browser removed the upload, the queue and the egress bill.
Read More
RustRust for Node Developers: What Finally Made It Click
I bounced off Rust twice. The third attempt worked because I picked a CPU-bound job: four minutes in Node became eleven seconds.
Read More
TypeScriptType-Safe Full Stack: Why TypeScript Alone Did Not Save Us
The API returned total as a string on one code path. The types said number. TypeScript was happy. A customer saw a shipping cost of "4995".
Read More
ReactReact Server Components: The Mental Model That Makes Them Click
A page shipped 340KB of JavaScript to render a read-only table. The cause was one 'use client' added months earlier, six levels up the tree.
Read More
AI AgentsAI Agent Memory: Why Your Assistant Forgets the Thing That Mattered
A user said they were in Germany in their first message. Fourteen turns later the assistant quoted UK pricing. Nothing failed — the fact had simply aged out.
Read More
Small Language ModelsSmall Language Models: The Cheapest AI Win Most Teams Skip
We were paying a frontier model to answer a yes-or-no question 6,000 times a day. Moving it to a small model cut the cost by 90% and one point of accuracy.
Read More
AI ObservabilityLLM Evals and Observability: Catching the Regression Nobody Sees
A classifier got eleven points worse and ran that way for six weeks. No error, no alert, no exception. A support lead mentioned it offhand.
Read More