July 7, 2026
Building an AI-powered CRM for community management
A product manager described building a fully functional CRM for a startup community entirely with Claude Code — roughly 57 hours of focused work without personally writing a line of code, at an AI cost of about 25 cents a month (the model estimated the same system would take a senior developer 600–1,100 hours). The design rests on a written "constitution" that always keeps a human in the loop, dual prose-and-compressed rule sets to keep the AI fast and accurate, and a three-layer validation system that auto-records hard facts, scores interpretations by confidence, and always pauses for human approval on personal data.
Local AI tools for cutting paperwork
A presenter showed how local AI tools (Codex and Claude for Work) reduce administrative overhead by keeping context in files within a local folder, with use cases such as extracting and categorizing invoices to save accountants time. Practical tips included always requesting interactive HTML for reports and CSV for data the model will re-read, "refreshing" an overloaded chat by exporting instructions to a Markdown file and starting a clean session, and using a browser extension to pull LinkedIn statistics into a CSV.
AI in PR, marketing, and media monitoring
A presenter shared how a university marketing team uses ChatGPT to cut board-report writing from two weeks to two days, categorizing raw media-monitoring exports into internal themes and comparing huge year-over-year course catalogs to flag changes. Other time-savers included asking the model to output finished HTML for the website, while always keeping final press-release edits in human hands to preserve tone, authenticity, and SEO performance.