💡 Use Cases
Real-world examples and success stories
Inside the $20.8K MRR Feature: 60 Prompts Over 14 Months on Claude
A tutoring platform built a session summary feature with Claude in 3 hours, then refined the prompt 60+ times over 14 months. The feature drives 22% of parent conversions and contributes to $20.8K MRR.
Practical Criticism of LLM Memory: Immutable Reflections & Ephemeral Sessions as Solutions
A critique of long-running sessions, life-companions, and LLM-wiki memory, offering solutions like immutable reflections, issue-bound session chains, and prompt templates to avoid intent loss and context overload.
Master Thesis Written Mostly with Claude: Economics Student Passes with Top Grade
An economics student used Claude for literature review, data analysis with Python scripts and Excel, and presentation — passed with a top grade near PhD level. Appendix disclosed AI use, no questions asked.
Building a Full App with Lovable + Claude + Gemini: A Case Study
A software engineer built Earnest, a bank bonus tracker, using Lovable for UI, Claude for intent understanding, and Gemini as a second opinion. The app now has 100+ users tracking $9,700+ in bonuses.
How One Team Replaced a 6-Figure HubSpot Agency with Claude Code
A mid-sized e-commerce company built their entire HubSpot Enterprise migration using Claude Code, replacing quotes of 20k-80k EUR for partial setups. They built 6 custom objects, 5 n8n integrations, and a KlickTipp migration in 4 months, with Claude Code handling both code and documentation.
Agentic Infrastructure: Replacing Splunk with Claude Code Agents for Server Monitoring
A developer deploys Claude Code sessions as services — router, monitors, dashboard poller — connected via WebSocket hub. Watchers are cheap bash; LLM wakes every 5 min for drain cycle. Dashboard tiles are natural-language queries cached in SQLite.
Enterprise AI agents: OpenClaw for channels, custom MCP tools, Cursor CLI runtime
Running AI agents in production for compliance, devops, and finance requires deterministic tooling, not raw API access. This post details a recipe: OpenClaw for channels, custom MCP per process, Cursor CLI as the agent runtime via ACPX, and self-hosted Kubernetes with immutable agent code.
Debugging a Pi Zero 2W BadUSB with Claude Code: Fixing an 'Impossible' Bug
A developer rebuilt a Pi Zero 2W BadUSB toolkit with Claude Code, which diagnosed a wrong-signal bug, empirically confirmed hardware limitations, and fixed a silent Python no-op in under 4 hours.
Claude for Engineering Compliance: 6-Month Workflow Breakdown
A technical firm shares how they use Claude Projects, Artifacts, and constraint-following to avoid hallucinations in client-facing specs.
Analyzing 7 Years of Diary Entries with an LLM: RAG vs Fine-Tuning Failures
After keeping a diary since 2019, a developer fed 200+ entries to an LLM to discover patterns — RAG failed, fine-tuning failed, and privacy was a constraint. The final approach revealed cyclical life lessons every two years.
Timezone Bugs in AI-Generated Booking Systems: A Case Study
Claude-generated booking prototype stored UTC, displayed in IST – broke for Dubai and US tutors. 11 wrong bookings cost trust. Fixed with luxon.
Accidental Dashboard Built with Claude Created a Product Commitment Nightmare
A developer built a dashboard with Claude in 2 days, forgot to feature-flag it, 40 customers found it and love it. Now customers want customization, requiring a 3-week refactor to make the hardcoded code extensible.