logicalbytes.dev - Engineering Blog
Building tools and systems that ship.
CI/CD Optimizations, Part 1: PR Checks
Two cache bugs, one lint flag, and a vulnerability database downloaded forty times a day. Fixing them cut Go jobs by 65% and PR time barely moved. Finding the job that actually set PR time took it from 5.3 minutes to 2.5.
Selected Work
The MCP Token Problem (and Three Ways to Solve It)
Most MCP servers burn 50-80K tokens before the user even asks a question. Here's why, and three generation modes that fix it.
Vigil: Producing the Receipts When Your ISP Won't
I built an open-source network reliability monitor because my building gates the ISP, my Slack huddles drop on certain afternoons, and the property manager keeps blaming my modem. I needed evidence.
Software Engineers Need Reckless Abandonment
After 15 years, I still don't know most things. That's the point. Be bold enough to dive into what you don't know, ask questions at every level, and push for what you believe in.
Pilum: From Launch to Production-Ready in 3 Months
I open-sourced a deployment CLI in December. Then I tried to actually use it to ship 4 products. Here's everything that broke.
Archive
- Symbiosisware: Building Software for an Audience of One Simon Tatham named the software you build only for yourself. AI makes it cheap enough that everyone is about to start building it.
- Capsa: Stop Renting Your Documentation I built an open-source documentation platform because the good managed option wanted ~$5k a year to host what is fundamentally a Markdown renderer. If someone else hosts your docs, you're renting your documentation.
- Building a Business at 2am It's 2am. My fiance is asleep, my cat is snoring, and I'm waiting for a release pipeline. Notes on building a business, not a startup, while the house gets framed wall by wall.
- The Difference Between Good Code and Bad Code Is 5 Minutes KISS applies to fixing systems too. Spend the extra 5 minutes to make it maintainable, and learn to spot when those 5 minutes are secretly 5 months.
- From Demo to Product: Telling the Story of Your Work Pressing a button speaks a thousand words. Rewrites, migrations, bug fixes: none of it matters if non-technical people can't see why.
- How I Start With Any Inherited Codebase Dropped into an unfamiliar codebase? The process I use to get the lay of the land on any project: Makefiles, architecture docs, and AI tooling.
- The Real Economic Threat Isn't AI — It's Automation at Scale The existential threat to labor isn't large language models. It's physical-world automation, and it has very little to do with chatbots.
- We're Not in an AI Bubble — We're Watching Big Tech Finally Spend Its Cash AI isn't a bubble. It's big tech finally deploying tens of billions in capital that sat idle for years.
- RAG vs NLP vs Raw LLM Calls: A Startup's Guide to Choosing the Right AI Architecture Startups should probably start with prompts. The honest tradeoffs between RAG, classical NLP, and raw GPT calls, and when each one actually earns its complexity.
- Introducing Pilum: A Recipe-Driven Deployment Orchestrator I built an open-source deployment orchestrator because I had a monorepo with Cloud Run containers, Cloud Functions, Lambda, and a frontend. I didn't want 5 different infra tools.
- Building Scalable APIs with Go: A Startup's Guide to Not Overengineering Unless you're building the next TikTok, scale isn't your problem. The math most startups ignore, when containers beat serverless, and how to avoid architecture astronaut syndrome.
- Managing Monorepos at Scale: Lessons from Meta, AWS, and a 15-Repo Migration What I learned about monorepos from Meta's massive codebase, AWS's polyrepo architecture, and migrating SID from 15 repositories to one.
Building
"A void in complexity is the signature of intelligence"