Loading…
Blog
RSS feedNotes on systems that have to stay up
152 long-form posts plus 28 LinkedIn articles on engineering, architecture, and the failure modes that only appear under load.
30 of 179
- 01Flash-Sale Ticketing: 2M Users, 500 Seats, No Double-BooksAt 10:00:00 sharp, two million people hit the same endpoint for five hundred seats. Your database is not the answer — it's the casualty. This is a full breakdown of the admission control layer, the atomic inventory counter, the hold lifecycle, and the payment reconciliation edge cases that actually cause double bookings in production.
- 02The Billable Hour Is Dead — AI Unbundled India's $315B IT IndustryTCS runs 80% of its business-services contracts on outcomes, five-year deal values shrank up to 35%, and a 500-person migration now prices at 50 people. The hourly billing model isn't being discounted — it's being taken apart layer by layer.
- 03AI Agents: A Practical Guide to Autonomous SystemsExplore the core components of AI agents, avoid common pitfalls, and implement a reliable system using proven patterns.
- 04You Don't Need Microservices. You Need Boundaries.Microservices won't fix a monolith that lacks boundaries — they'll just move the coupling onto a network. Break a Laravel app into Billing, Catalog, Identity, and Reporting modules with owned data, public contracts, and Pest-enforced dependency rules, and only extract to a service when a measured problem demands it.
- 05The Agentic Loop: When AI Should Think, Act, and LoopEvery AI system you rely on — scripts, watchdogs, coding agents, Grok bots — runs the same core loop: observe, decide, act, verify. This post breaks down that loop, the two ways to put a human around it, and a practical framework for deciding when an agentic loop earns its cost and when a plain script wins.
- 06How HTTPS and TLS Actually Work: A Deep Dive into Secure Web ConnectionsA detailed walkthrough of how TLS creates secure web connections using Diffie-Hellman key exchange, certificates, and authentication to protect your data on the public internet.
- 07The AI Employee Era: Build Your Own AI TeamA practical roundup of this week's biggest AI moves — from AI bots that act like real teammates to coding models that undercut the leaders — plus a step-by-step guide to building your own AI employee in minutes.
- 08Building Enterprise AI Agents on Google Cloud: Identity, Traces, and TrustDeploying AI agents isn't just about code—it's about identity, audit trails, security, and memory. This post breaks down Google's Gemini Enterprise Agent Platform and the four critical systems you need for production-ready agents.
- 09The Missing HTTP Method: QUERY Is Almost HereComplex queries don't fit in GET URLs, so we abuse POST and lose caching, safety, and automatic retries. The HTTP QUERY method — now in the RFC Editor queue as a Proposed Standard — fixes this. This deep dive covers the spec's semantics, caching rules, the Accept-Query header, and practical implementation patterns for Laravel and Node.js.
- 10Laravel & PHP Guidelines for AI Code Assistants — A Practical Style GuideAI assistants generate Laravel code fast, but rarely in a consistent house style. This post distills a battle-tested set of Laravel & PHP conventions — typed properties, happy-path-last control flow, naming rules, and testing structure — into a guide you can drop straight into your AI tooling. Inspired by Spatie's well-known guidelines.
- 11Why AI Hasn't Replaced Engineers — And Won't Anytime SoonAI writes most of the code now, layoffs are running at over 1,000 a day, and execs are saying they hired zero new engineers this year. Yet companies are quietly rehiring the same people they let go, agents that score 80% on SWE-bench drop to 12% on real industrial tasks, and the scarce resource has shifted from typing to judgment. This is a data-backed breakdown of why the discipline survives even as the job changes.
- 12Circuit Breakers and Bulkheads: Containing FailureOne slow dependency can consume every thread in your gateway and take down a perfectly healthy API. This post breaks down the two patterns that contain the blast radius — circuit breakers and bulkheads — what each actually solves, why they're different, and how to wire both with Resilience4j.
- 13CRDTs: How Google Docs Survives Four People Typing at OnceMulti-user editing looks smooth, but underneath it's a distributed systems nightmare of shifting indexes. This post breaks down how CRDTs make merge conflicts mathematically impossible — character IDs, origin pointers, tombstones, and the YATA algorithm behind Yjs.
- 14System Design Fundamentals: The 10 Components Behind Every App That ScalesAnyone can write code that works. System design is what makes it work for 100 million users at once. This post walks through every core building block — databases, caching, load balancing, replication, partitioning, CAP theorem, message queues, and monitoring — with fact-checked numbers (WhatsApp moves 100+ billion messages a day, not millions) and a worked video-streaming design at the end.
- 15The Ticking Clock: Why Short-Lived JWTs Are Your Best DefenseStateless JWTs can't be revoked on demand — once issued, they live until they expire. This deep dive covers why short lifetimes (5–15 min) are the strongest control you have, how the access/refresh token split actually works, refresh token rotation with reuse detection, and the sender-constrained token guidance from RFC 9700 (Jan 2025).
- 16Tokens Are Compute — Why Your LLM Bill Is Really a GPU BillMost engineers reason about LLMs in words, characters, or messages. The model sees none of that — it sees tokens, and tokens are compute someone's GPU has to run. This post traces what a token actually is, why output costs 3–10x more than input, the five-step journey of an API call, and the four cost levers most teams never pull.
- 17Backend-for-Frontend (BFF): The Missing Piece in Your Fullstack AppYour frontend is doing too much — orchestrating microservices, parsing messy payloads, and worst of all, holding auth tokens the browser can't keep safe. The Backend-for-Frontend pattern moves all of that to a server layer you control. Here's why it matters (especially after the September 2025 npm attack that hit packages with 2.6B weekly downloads) and how Next.js gives you one for free.
- 185 WebSocket Design Decisions Most Developers Never LearnMillions of developers ship chat apps, live dashboards, and multiplayer games on WebSockets without understanding the protocol's hidden design choices. This deep-dive unpacks five of them — the magic GUID handshake, TCP slow-start exploitation, client-side masking, firewall-friendly ports, and why HTTP/2 sometimes wins. You'll walk away able to explain the protocol at a level that holds up in any system design interview.
- 19PageIndex: Vectorless RAG That Reasons Through DocumentsTraditional RAG chops documents into arbitrary chunks, embeds them, and hopes cosine similarity finds the right one. PageIndex throws that out — it builds a hierarchical table-of-contents tree and lets an LLM reason its way to the right section, the way a human expert flips to the right chapter. No embeddings, no vector DB. It hit 98.7% on the FinanceBench benchmark.
- 20MCP 2.0: The Stateless Rewrite That Breaks Every Server You've ShippedThe next MCP spec drops session IDs, removes the initialize handshake, and turns Extensions, Tasks, and MCP Apps into first-class citizens. It's the largest revision since launch — and it contains breaking changes. Here's every change, why it exists, and what you have to migrate before July 28, 2026.
- 21Time-Series Databases: Billions of Points, Bytes EachTraditional B-tree databases collapse under append-only telemetry. This is a deep dive into how TSDBs ingest millions of points per second — WAL, MemTables, LSM compaction, and the Gorilla compression that squeezes a 16-byte data point down to ~1.37 bytes. Covers cardinality, data modeling, and how to pick between Prometheus, VictoriaMetrics, TimescaleDB, and InfluxDB.
- 22Infrastructure as Code Solved Repeatability. It Didn't Solve Coordination.Terraform fixed the snowflake-server problem. But once you have dozens of stacks, multiple teams, and dependencies between environments, a new class of failure shows up: drift, ordering, and compliance gaps. This is a deep dive into why provisioning isn't orchestration, and the control-plane pattern the best teams build on top of their IaC.
- 23Your Site Looks Fine — Until It Loads on a PhoneMost slow sites aren't badly built — they just do too much work to load. This guide walks the critical rendering path step by step, then fixes the four things Lighthouse always flags: CSS, images, JavaScript, and fonts. Includes the delivery layer (CDN, HTTP/3, Brotli, caching) and how to stop performance from eroding after you fix it.
- 24Inside Instagram's Recommendation SystemThis week a 4-day-old account crossed 10 million followers and overtook a major political party. The follower count is the headline, but the real story is the architecture underneath. We break down how Instagram went from a chronological photo stream to a four-stage recommendation funnel — fan-out, two-tower retrieval, multi-stage ranking, and cold start — and what every backend engineer can steal from it.
- 25ReBAC Explained: How GitHub Decides Who Can Touch Your RepoRoles work until your app grows hierarchies, then you drown in role explosion and token bloat. Relationship-Based Access Control (ReBAC) answers 'can this user do this?' by walking a graph of relationships instead. This post explains ReBAC using GitHub's permission model as the main example, then covers Google Zanzibar — which manages over 2 trillion relationship tuples at 10 million queries per second.
- 26Why Google Built Two AI Chips Instead of One — Inside the TPU v8 SplitGoogle's eighth-generation TPU isn't a chip — it's two. The TPU 8t targets training, the TPU 8i targets agentic inference, and the reason for the split tells you where AI hardware is heading. This post builds the story from the silicon up: how CPUs, GPUs, and TPUs actually differ, what a systolic array does, and why one chip can no longer be optimal for both halves of the AI workload.
- 27The GitHub Exodus: Why Developers Are Walking Away in 2026GitHub uptime has collapsed to roughly 86% in April 2026, 292 pull requests vanished from a Merge Queue bug, and Mitchell Hashimoto — user #1,299 — has packed up Ghostty and left. Here's what's actually happening, why AI agents are part of the story, and where developers are migrating.
- 28Kafka vs RabbitMQ vs SQS — Stop Picking the Wrong QueueMost outages in queue-based systems aren't tool problems — they're concept problems. Pub/sub vs point-to-point, delivery guarantees, dead letter queues, and back pressure decide whether your system survives a bad Monday. This deep dive covers all four, then walks through when Kafka, RabbitMQ, or SQS is the right call — with real throughput numbers and the 'exactly-once' myth busted.
- 297 API Architectures Every Developer Should Know in 2026Picking the wrong API style costs you in latency, bandwidth, and 3 AM debugging sessions. This post breaks down the seven API architectures powering modern apps — what they do, where they shine, and the trade-offs you need to know before you commit.
- 30Why SpaceX Is Paying $60B for CursorOn April 21, 2026, SpaceX announced an option to acquire Cursor for $60 billion later this year. This post unpacks the actual architecture that makes Cursor feel like a senior engineer — local indexing, Merkle-tree change detection, Tree-sitter AST chunking, Turbopuffer vector search, and the Composer agent model — then breaks down the three strategic reasons this deal is really about who owns the AI stack.