April 23, 2026
Good talking with you today. You mentioned you're interested in building with AI — and you're already building apps for nursing homes to cut down on manual work. This page is a quick look behind the curtain of how I work day-to-day with Claude Code, plus a handful of concrete ideas for your nursing home vertical.
Jump to nursing home ideasClaude Code is not a code generator you prompt and babysit. It's an orchestration layer. I describe what I want, break it into tasks, and dispatch specialized agents to handle each one while I stay focused on what's next.
Describe the goal in plain English. Claude breaks it into tasks and confirms scope before touching a file.
Hooks enforce TDD. A pre-commit guard refuses to proceed if tests don't exist or don't pass.
Multiple subagents run at once — one writes the feature, another reviews it, another deploys.
Deployed. Claude logs a memory of what worked so future sessions don't re-figure the same thing.
Haiku handles obvious, single-file edits. Sonnet handles judgment calls across multiple files. Opus handles architectural decisions. The main context stays clean — it only orchestrates, never does the work directly.
Every project has a SESSIONS.md with recent decisions and conventions. A hook injects
relevant learnings into every new session automatically. I never re-explain the same thing twice.
PostToolUse and Stop hooks run the test suite after every file change. If tests fail, the agent pauses. A stop-guard uses a Haiku gate to skip tests during planning — and runs them during coding.
Cloudflare Pages for static sites. Docker Swarm on a home server for stateful apps. Traefik for routing. SpacetimeDB for real-time. Laravel + Vue for traditional web. Git commit-to-main, no ceremony.
A full pitch page like this one — researching reference designs, writing all the copy, building the HTML, deploying, DNS, smoke tests — runs in about 20 minutes. Not because I'm fast. Because most of it runs in parallel while I'm doing something else.
You mentioned you're building small applications to reduce manual work for nursing home staff. These are five specific ideas grounded in the real pain points of that environment — each one you could build yourself with Claude Code, or we could build together.
"CNAs dictate their end-of-shift notes. AI structures them, flags what changed, routes alerts to the incoming RN."
The shift handoff is one of the riskiest moments in a care facility — information gets lost, verbal reports get rushed, and critical changes since the last shift slip through. Right now most CNAs type or handwrite notes under time pressure at the end of a 12-hour shift. Voice dictation + AI structuring turns a 20-minute documentation grind into 90 seconds of talking while walking out.
"Nurses speak resident updates room-to-room. AI parses them into structured ADL and mood records."
Activities of Daily Living documentation is mandatory, repetitive, and hated. Staff write the same information twice — once on a clipboard, once in the EHR. A phone mic + Claude turns the clipboard step into the EHR step.
"Weekly plain-English family updates drafted from care notes, approved in one click."
Most facilities have a standing expectation to communicate with families weekly. A director of nursing often spends Sunday afternoon writing personalized updates. This generates a draft from existing care notes with a single nurse-approval step before sending.
"Scans progress notes for missing required fields before the state surveyor shows up."
State surveys can cost a facility thousands in citations for documentation deficiencies that are entirely preventable. The auditor scans notes against state-specific survey rules and flags gaps with enough lead time to fix them. Most facilities have zero tooling for this.
"Grounded in the facility's own handbook. New hires ask questions, get correct answers, stop interrupting charge nurses."
Onboarding turnover in nursing homes is brutal. New CNAs and aides interrupt charge nurses constantly with questions that are answered in the handbook — which nobody reads fully. A RAG-based chat grounded in the facility's own policies handles the 80% of questions that are already answered somewhere.
Three modes, depending on how you want to work. No pitch here — just options.
You're already building. I can be a thinking partner — reviewing what you're building, helping you think through AI integration, and showing you the Claude Code patterns that'll save you the most time in your stack.
Pick one of the ideas above. We run a focused build session — a few hours, a working prototype in your hands. A real proof-of-concept you can take to a facility for feedback, not a mockup.
The Exponential Advisor work your dad is building is about exactly this kind of leverage. If you want to explore what a nursing-home-focused AI product looks like at real scale, that's a conversation worth having together.
Nursing home staff are some of the most overworked, underpaid people in healthcare. Tools that cut even 20 minutes of paperwork per shift make a real difference in their days. That's worth building well.
Reply to any email, text, or just hit the button below. Happy to jump on a call, share code, or just compare notes on what you've been building.
ethan@braceyourself.solutions
This page was built the same way I described — designed, written, deployed, and DNS'd in about 20 minutes
using Claude Code. The source lives at ~/relationships/johnanderson2/index.html
if you ever want to poke around.