Writing
What we learn building and operating our own software, and helping teams put AI into production. No hot takes — just what actually worked.
A build log for a static Astro site that hits 99 Performance, 100 Accessibility, LCP 2.0s and under 2KB of JavaScript — self-hosted fonts, hard-offset shadows, and no framework on the client.
GummySearch shut down over Reddit API licensing, leaving 135,000 users looking for a replacement. A comparison of F5Bot, Syften, Brand24 and RedNudge — on price, on scoring, and on the structural risk that killed GummySearch.
ShipFast, Makerkit, Supastarter and SaaSInMinutes compared on price, what's actually in the box, and the one thing most boilerplates leave you to figure out — getting it deployed on a real domain.
A comparison of Taplio, AuthoredUp, Supergrow and Propelr on the two axes that matter — what you pay per month, and what the tool does to your LinkedIn account to do its job.
Every SaaS begins with the same 40 hours of undifferentiated plumbing — auth, payments, email, monitoring, deploy. Here's the full checklist, and why building it from scratch each time is a tax.
Why LinkedIn automation tools that use browser extensions and page scraping put your account at risk — and why the official Marketing Developer API is the safer foundation.
Why starting is the hardest part for ADHD brains, why to-do lists make it worse, and how to design software for task initiation instead of task management.
How to monitor brand and competitor mentions on Reddit without living on the site — keyword tracking, relevance scoring, and the daily-digest workflow that catches high-intent threads.
The specific tells that mark text as machine-written — and why voice-trained generation beats generic prompting for content that actually sounds like a person wrote it.
The unglamorous engineering that turns an AI demo into a production feature: evals, fallbacks, cost controls, prompt versioning, and the human checkpoints that keep it safe.
The systems, stack, and hard rules that let one person build and operate four live products — Inchworm, Propelr, RedNudge and SaaSInMinutes — without a team or burning out.
A practical framework for evaluating AI use cases before you build: the four questions that separate projects that ship value from demos that die in a slide deck.