Tickets That Are Tests
How xswarm builds xswarm: every ticket is a Gherkin scenario written first, and the pipeline codes, tests, stages and ships it.
Read Full ArticleBlog
Brutally honest insights on AI development, token economics, and building with swarms of specialized agents.
RSS
Why Your AI Assistant Can't Remember Where Anything Goes
We've all been there - Claude Code writes brilliant functions but can't figure out where they should live. What if that AI intern came with a full senior team?
Read Full Article
AI Development: Lessons Learned
Lessons learned working with AI coding assistants: practical patterns and solutions that emerged from real-world experience
Read Full Article
The Day I Realized My AI Assistant Was Working Like a 1995 Modem
AI coding tools still work one task at a time, like CVS and waterfall. Parallel agent streams remove that bottleneck.
Read Full Article
I Found 15 Different Email Validation Functions in My AI-Generated Code
AI-generated codebases rewrite the same utility functions over and over. Extracting pure functions during planning stops it.
Read Full Article
How My AI 'Optimized' Auth Middleware to Give Everyone Admin Access
"It works in my AI sandbox" is the new "it works on my machine." Full mock environments let you integration-test before deploy.
Read Full Article
What Happened When We Stopped Showing Stakeholders PDFs
Stakeholders still say "make it pop" about static PDFs. Deploying working prototypes to GitHub Pages lets them click real interfaces.
Read Full Article
The Sprint Demo Where Marketing Asked for Screenshots We Never Took
You ship an AI-built feature and marketing wants screenshots, demos and posts. Generating those assets during development ends the post-launch scramble.
Read Full Article
I Inherited a Codebase Where AI Wrote Everything. Here's What I Found.
AI piles up technical debt because it doesn't know your architecture. Multi-pass planning makes it build on what exists instead of duplicating it.
Read Full Article
Why I Stopped Writing Perfect Prompts and Started Deploying AI Teams
Multi-agent orchestration ends repetitive prompting by splitting work across specialized AI teams. Epoch cycles let complex tasks run in parallel.
Read Full Article
Our 15-Minute Standups Became 3 Minutes When AI Teams Don't Argue
AI teams report through structured status updates from isolated workspaces, so nothing gets crossed. A live Kanban board shows every agent at a glance.
Read Full Article
Sequential Development Died When We Started Pair Programming With Entities That Never Sleep
AI development today is chaos: context switching, endless prompts, scope creep. AI Epochs are full build cycles shaped for AI, not human limits.
Read Full Article
I Discovered My AI Assistant Was Actually 40 Specialists Pretending to Be One
Beyond the coding assistant sit specialist AI agents for QA, security testing, PR management and content. Here are 40+ roles and how each works.
Read Full Article
Tracking Token Usage in AI Development: Lessons from Real Projects
Where tokens go in AI-assisted development, and how better context management and code reuse cut the waste.
Read Full Article