Blog.
Thoughts on AI agents, software development, and building in public.
Engineering
Encrypted against you, plaintext to your cheaper model
Researchers extracted 315,320 reasoning blocks from public agent logs using a decryption oracle attack. 182 credentials leaked. The encryption was real. The key management was not.
Engineering
The Permission Prompt Was Never Going to Save You
Anthropic made Claude Code auto mode the default. The permission prompt was already broken. But all the safety data comes from the vendor.
Engineering
The Swarm Myth
Multi-agent AI systems degrade performance 39-70% on sequential reasoning. The coordination tax, failure cascades, and why one agent with good tools wins.
Security
AI Wrote the Bugs. Now AI Fixes Them.
AI-generated code has 2.74x more security flaws than human code. OpenAI's new Codex Security CLI finds and patches them. The circular quality loop is now a product.
Engineering
The Token Bill Comes Due
Uber burned its 2026 AI budget by April. Microsoft pulled Claude Code. A $500M month. AI coding tools hit production scale before anyone built the financial controls. The bill is here.
AI
The $165K Rewrite
One engineer rewrote Bun from Zig to Rust in 11 days. 1M lines of code, $165K API bill, 64 Claude instances. What the numbers actually prove about AI-assisted engineering at scale.
AI
Can You Trust Your Team's Commits?
repo-slopscore scans 5,337 git repositories for AI-generated code. The tool is imperfect. The question it raises about code provenance is not.
AI
The Browser Is the New API
Apple ships a Safari MCP server. When the most closed platform vendor adopts your open protocol, the browser becomes an agent-native surface.
AI
When the Code Reads You Back
Claude Code silently embedded invisible Unicode steganographic markers into system prompts. A trust story about AI coding tools and the transparency bar they must meet.
AI
The Agent Loop Is the New Prompt Engineering
How self-scaffolding RL in Ornith-1.0 moves agent loop design from manual prompt engineering into model training.
AI
A $130 Charge for Code Nobody Wrote
Cursor users report surprise on-demand charges and locked accounts. The billing transparency crisis in AI coding tools is reaching a tipping point.
AI
SpaceX Just Bought Cursor for $60 Billion. Your Code Is the Product.
SpaceX bought Cursor for $60B. One million developers' code now flows to a company under active AI safety investigation. Here's what changes.
AI
Stack Overflow Built a Platform for AI Agents. Its Community Wants It Gone.
Stack Overflow launched an API-first knowledge exchange for AI agents. Its community downvoted it into the ground. The problem it solves is real. Whether Stack Overflow is the right entity to solve it is the open question.
AI
Anthropic Built the Most Powerful Coding Model Ever. Then They Secretly Weakened It.
Claude Fable 5 can migrate 50 million lines of Ruby in a day. It can also silently decide you don't deserve its best answer — and not tell you. The story of the most capable public AI model ever shipped, and the transparency crisis it triggered within 48 hours.
AI
AI Writes Most of Your Code Now. That's the Problem.
AI coding tools shipped 66% more epics per developer. They also tripled production incidents, spiked bugs by 54%, and buried senior engineers in review queues. The Faros AI Engineering Report 2026 reveals what the dashboards miss.
AI
AI Writes 42% of Your Code. Why Haven't You Shipped 42% Faster?
AI writes 42% of committed code. Teams are not shipping 42% faster. The data explains why.
AI
The Quiet Death of the Code Review
AI code review tools are quietly replacing the way engineering teams handle pull requests. The data shows adoption growing 14x in 21 months, but developers trust AI code less than ever.
AI
Composer 2.5 is the AI coding model most developers should actually use
At $20/month all three major AI coding tools cost the same. The difference is how far that budget goes. Composer 2.5 stretches your credits further than any competitor.
AI
AI Is Creating a Two-Tier Engineering Economy
AI coding tools split engineering into visible shippers and invisible reviewers. 94% of leaders admit their metrics cannot tell the difference.
AI & Design
A Jane Street Designer Stopped Using Figma
Edwin Morris at Jane Street replaced Figma with Claude Code for internal tool prototyping. The three paths ahead for designers in an AI-first world.
AI
87% of Devs "Feel Faster." The Data Is a Mess.
The 55% faster statistic is everywhere. But the studies behind AI coding productivity tell a far messier story.
Software Engineering
The Quiet Death of the Code Review
AI code review tools aren't replacing humans — they're creating a three-tier system where juniors learn from bots and seniors focus on architecture. The metrics haven't caught up yet.
Career
Your 10 Years of Expertise Is Now Promptable
A senior engineer's three pillars — domain knowledge, debugging intuition, architectural judgment — are dissolving into prompts. The economics of homogenization.
AI Engineering
Loop Engineering on a $1.3M Budget
Peter Steinberger spent $1.3M on autonomous agents. Boris Cherny says he doesn't prompt anymore. Here's how the rest of us build loop engineering on real budgets.
AI
The $20 AI Coding Bet: Claude Code vs Codex CLI
Two terminal agents, one price point, wildly different economics. I've been running both Claude Code and Codex CLI on the $20 tier for the past month, and the r
DeFi
YieldRadar: Building a DeFi Yield Intelligence Bot
How I built a multi-factor risk scoring system for 18K+ DeFi yield pools, from scratch to production Telegram bot.
AI
Claude Code vs Cursor in 2026: The Token Efficiency Gap
Why the cheapest option isn't always the one with the lowest sticker price.
AI
From Vibe Coding to Deliberate Design: My First Week with Kiro IDE
The AI coding landscape is finally growing up. Kiro's spec-oriented development turns chaotic AI coding into a predictable, accountable collaboration.
AI
The Great AI Coding Shake-Up of 2025: Why I'm Breaking Up with Windsurf to Augment Code
A story from the trenches about breaking up with Windsurf, leaving Cursor behind, and finding a new AI co-pilot in Augment Code.
Career
Beyond Slack Emojis: Mastering Empathetic Asynchronous Communication as a Developer
(Because 'LGTM 👍' Doesn't Always Cut It) — Mastering empathetic async comms isn't just fluff; it's a core skill for saving time, reducing friction, and building awesome software together.
AI
Ditch the AI Glue Code: Meet MCP, the Universal Connector to Your Dev Workflow Needs
An open standard called Model Context Protocol (MCP) is the 'USB-C port for AI' — letting models seamlessly talk to databases, APIs, and developer tools.
AI
Confessions of an AI Whisperer: Dodging Pitfalls and Mastering the Art of the Prompt
A developer's field guide to prompting — common blunders, practical mastery techniques, and why the human element still matters most when working with AI.
AI
Copilot vs. Cursor vs. Windsurf: Which AI Coding IDE is Right for You?
A three-way showdown between GitHub Copilot, Cursor AI, and Windsurf IDE — comparing strengths, weaknesses, pricing, and the best use case for each.
AI
Decoding the Magic: A Deep Dive into AI Agents
AI agents are the superheroes of the AI universe — autonomous, adaptable mini-masterminds that can think, act, and learn on their own.
Career
Beyond the Code: Why Being a "Jack of All Trades" (and Master of One) Makes You a Tech Unicorn
Becoming a T-shaped professional — deep expertise in one area plus broad knowledge across the tech landscape — is the key to a thriving, future-proof career.
Career
Lessons From the Tech Trenches: 5 Things I Wish I Knew When I Started
Five hard-won lessons from 5+ years in the tech jungle — consistency over scattered brilliance, embracing the struggle, and knowing when to jump ship.
TypeScript
Coding Commandments: 10 Clean Code Rules Even Your Rubber Duck Will Approve Of
Ten practical clean code commandments inspired by Uncle Bob Martin — naming, dead code, tests, refactoring, and keeping it simple.
Career
How to Sleep Like a Baby (Even When You're Debugging Code Nightmares) — A Developer's Guide to Catching Z's
A developer's guide to prioritizing sleep — daytime prep, pre-bedtime rituals, and nighttime ninja moves for when sleep resists capture.
Career
10 Ways to Stay Motivated as a Remote Worker (a.k.a. How to Avoid Becoming a Hermit in Your Pyjamas)
A survival guide packed with tips for staying motivated while working remotely — from creating a Batcave to embracing the virtual water cooler.
TypeScript
TypeScript: Rise of the Phoenix — Node.js Embraces the Uncompiled Future (and SWC Joins the Party!)
Node.js adds experimental native TypeScript support via type stripping and SWC integration — eliminating the compile step and unifying the JS/TS ecosystem.
Career
Feedback Loops and Rubber Duckies: A Dev's Guide to Not Taking it Personally (and Actually Improving)
How to give and receive feedback without taking it personally — practical tips for code reviews, swallowing your pride, and turning critique into growth.
TypeScript
Authentication Showdown: Session Cookies vs. JWT vs. PASETO — Which One Reigns Supreme?
Comparing three popular authentication methods — session cookies, JWT, and PASETO — across security, scalability, and developer experience.
Career
Why Your Soft Skills Matter More Than You Think (And It's Not Just About Being Nice)
Soft skills are the social glue that holds teams together. Communication, collaboration, and emotional intelligence can make or break your developer career.
TypeScript
Code Comments: Your Secret Weapon for a Happier Dev Life (and a Healthier Codebase)
How to write code comments that actually help — TODOs, FIXMEs, inline explanations, and the anti-patterns to avoid in your codebase.
TypeScript
Branded Types in TypeScript: Stronger Type Safety for Your Codebase
TypeScript is fantastic for catching type-related errors at compile time. However, there are scenarios where its structural type system falls short. This is where branded types step in.
Career
The Remote Work Revolution: Embracing Flexibility and Work-Life Balance
Remote work offers flexibility and work-life balance that traditional offices can't match — especially valuable for developers juggling career and family.
TypeScript
Promises in JavaScript: When to Reconsider Promise.all()
JavaScript's Promise.all() offers a tempting way to handle multiple asynchronous tasks at once. However, it's crucial to recognize its potential shortcomings in real-world development scenarios.
TypeScript
PNPM: The Smarter, More Efficient Node.js Package Manager
NPM has ruled the Node.js dependency management landscape for years. But as projects grow complex, NPM's limitations become increasingly apparent. Enter PNPM.