A higher-ed technologist spent a year building custom software with Claude Code, shipping 53 working tools from 76 ideas across 172,486 lines of code. Here's what stuck and what didn't.
A higher-ed technologist spent a year building custom software with Claude Code, shipping 53 working tools from 76 ideas across 172,486 lines of code. Here's what stuck and what didn't.
A sysadmin spent 6 hours using AI to write a Bash DHCP script and documented every regression, hallucination, and loop. Here is what the AI itself admitted went wrong.
A developer spent ten days building a Mac list manager with Claude Code, then burned an entire weekend on one intermittent bug. Here is what the actual numbers reveal about vibe coding at scale.
Five fleet professionals are using vibe coding and AI to build custom fleet tools in-house. One cut an 80-hour task to 30 minutes. Here is how they did it.
Ohalo Genetics CEO David Friedberg tested AI-built horizontal tools before landing on a buy-vs-build framework. The lesson applies to any founder running a lean stack.
Best Lawyers CEO Phillip Greer built an internal AI app store on top of a unified data layer. Now 40% of non-engineers ship dashboards and automate reports. Here is how they did it.
A developer built three syncing personal apps using Claude on an £18/month plan. Here is the architecture, the process, and the honest tradeoffs.
A South American gift company moved its upsells off the product page and onto a dedicated step. Order value jumped 28%. Here is the methodology behind the test.
One developer ran Claude Code and ChatGPT Codex on the same codebase simultaneously. This is the instruction file architecture that kept them from contradicting each other.
A Silicon Valley engineer went from Claude Code convert to near-termination after AI-generated bugs crashed production twice. Cal Newport unpacks what went wrong.
A CRO agency removed a funnel stage for a major weight-loss subscription company and revenue jumped 26%. Here is the research logic and what made it work.
A Thoughtworks CTO ran a controlled experiment: refactor a 17,155-line file, then measure token costs for identical changes. Input tokens dropped from 159,564 to 27,360.