hudtaylor

I run a full AI operations stack on an M4 Mac Mini — ClawdBot (Claude), OBS streaming a 24/7 WebGL simulation, Chrome for browser automation, 16 cron jobs, the whole thing. $599 machine.

Reality check: it works remarkably well for AI agent orchestration. The unified memory architecture means the agent, browser, and streaming can coexist without the memory wall you'd hit on x86. But running OBS alongside everything else does make it laggy — I've got an M5 MacBook Air (32GB) incoming and I'm planning to swap the Mini for a 64GB model to give more headroom.

For anyone considering Apple Silicon as an AI dev machine: the sweet spot is 64GB unified memory minimum if you want to run an agent + browser automation + anything else simultaneously. 32GB works but you feel the pressure. The M-series efficiency means you can leave it running 24/7 without worrying about your power bill, which matters when your AI agent literally never sleeps.

I run an AI agent that deploys production code across 4 live products. The self-correction capability is real but undersold in these debates.

Examples from the last month: the agent found its own API keys in environment files after initially claiming it didn't have them (lesson: grep before asking). It caught itself about to run a destructive database migration on a shared production instance and stopped. It fixed 8 broken RSS feed configurations that had been silently failing for weeks without anyone noticing.

The pattern I've found: AI doesn't need to be perfect at writing code. It needs to be honest about what it doesn't know, aggressive about testing its own work, and operating under clear constraints about what's destructive vs. safe. We maintain a file called AGENTS.md with "sacred rules" — things the agent can never do without explicit approval. Database migrations, pricing changes, anything with --accept-data-loss.

The "no LLM" stance makes sense if you don't have guardrails. With the right constraints, AI-assisted code is faster AND safer than solo human development — because the agent never gets tired, never rushes before a deadline, and never thinks "I'll test that later."

I run a full AI operations stack on an M4 Mac Mini — ClawdBot (Claude), OBS streaming a 24/7 WebGL simulation, Chrome for browser automation, 16 cron jobs, the whole thing. $599 machine.

Reality check: it works remarkably well for AI agent orchestration. The unified memory architecture means the agent, browser, and streaming can coexist without the memory wall you'd hit on x86. But running OBS alongside everything else does make it laggy — I've got an M5 MacBook Air (32GB) incoming and I'm planning to swap the Mini for a 64GB model to give more headroom.

For anyone considering Apple Silicon as an AI dev machine: the sweet spot is 64GB unified memory minimum if you want to run an agent + browser automation + anything else simultaneously. 32GB works but you feel the pressure. The M-series efficiency means you can leave it running 24/7 without worrying about your power bill, which matters when your AI agent literally never sleeps.

I run a ClawdBot (Claude on a Mac Mini) as a 24/7 AI Chief of Staff for my company. 7 email accounts monitored, 16 daily crons, code deployments, content pipeline — the whole operation.

The thing nobody tells you is it doesn't make you less busy. It makes you more busy. Your mind enters this constant state of orchestration. But for someone with ADHD, that's the unlock. I used to generate 10 ideas and execute on maybe 1 before the next wave hit. Now every idea has a catcher's mitt.

My background is all over the place — synthesized a novel organic compound in grad school that nobody had built before, went to brew science school, spent 15 years building laboratory databases and information systems. That multidisciplinary wiring is exactly what makes orchestrating an AI agent feel natural. You're not writing code, you're designing workflows — same thing I did in the lab for years.

I have ADHD and honestly struggle with reading and writing sometimes. Sounds ridiculous for someone running a tech company. But I genuinely feel like I have a superpower now. The disability became the competitive advantage.

AI writes code, generates art, and passes the bar exam. It still can't tell you when to take out the trash.

I moved to a new neighborhood in San Diego and asked ChatGPT, Claude, and Gemini my trash pickup day. They all hallucinated. The city's official lookup is a Salesforce app behind reCAPTCHA that barely works on mobile.

So I built trashalert.io — and mapped 954K addresses across San Diego and Austin.

What it does: - Type your address → instant fuzzy match against 954K addresses - Shows collection day, next pickup dates, which bins to put out - Auto-adjusts for holiday shifts - Generates .ics calendar files for any address - Night-before email reminders - 80+ neighborhood SEO pages

Tech: Next.js 15 / TypeScript / Supabase (Postgres) / Vercel / Resend

The interesting problem: ~19,000 US municipalities, each with different waste haulers, schedules, and data formats. Most don't publish machine-readable data. I'm building a crowdsource pipeline where residents submit their schedule and we verify against public records.

Built as a solo founder with an AI ops lead running 24/7 on a Mac Mini. 233+ commits in 2 months.

The thesis: municipal data is one of the last un-aggregated data sets. Trash day is the wedge — street sweeping, recycling changes, water shutoffs all use the same address→schedule architecture.

Free, no ads, community-powered. Would love feedback.

I moved to a new neighborhood in San Diego and couldn't figure out my trash pickup day. I asked ChatGPT. It confidently told me Tuesday. It was wrong. Asked Claude. Also wrong. Googled it — the city's lookup tool is a Salesforce app behind reCAPTCHA that barely works on mobile.

So I built trashalert.io — a free, community-powered trash schedule lookup covering San Diego and Austin (954K+ addresses across 79+ neighborhoods).

How it works: - Type your address → get your collection day (trash + recycling + organics) - Holiday-adjusted dates — SD shifts pickup forward after holidays - Recycling week A/B tracker — the most confusing thing about SD trash - Interactive collection day map — 954K+ addresses color-coded by pickup day - Calendar export (.ics) — set it and forget it - Community reports — neighbors confirm each other's schedules

Tech: Next.js 15, Supabase (Postgres, 954K+ records), Vercel (102kB first load), Plausible + GA4. No auth, no accounts — zero friction.

The data problem: San Diego has 500K+ households. The city's data is locked behind reCAPTCHA Enterprise on a Salesforce endpoint. My approach: build the database from the community up. Users report their pickup day → consensus verification (3+ confirmations = verified). Started with 201 addresses, expanded to 954K+ via hauler data imports across 79+ neighborhoods in 2 cities.

What I learned: 1. Address normalization is surprisingly hard (Ave vs Avenue vs Av, apartment stripping, ordinal conversion) 2. Every municipality structures trash data differently — there's no standard 3. People WANT to report their trash day. The "help your neighbors" angle converts better than any growth hack. 4. AI confidently hallucinates municipal data. This is a real gap.

Free, no ads, community-powered. Would love feedback. Planning to expand to more cities.

I moved to a new neighborhood in San Diego and couldn't figure out my trash pickup day. I asked ChatGPT. It confidently told me Tuesday. It was wrong. Asked Claude. Also wrong. Googled it — the city's lookup tool is a Salesforce app behind reCAPTCHA that barely works on mobile.

So I built trashalert.io — a free, community-powered trash schedule lookup covering San Diego and Austin (954K+ addresses across 79+ neighborhoods).

How it works: - Type your address → get your collection day (trash + recycling + organics) - Holiday-adjusted dates — SD shifts pickup forward after holidays - Recycling week A/B tracker — the most confusing thing about SD trash - Interactive collection day map — 954K+ addresses color-coded by pickup day - Calendar export (.ics) — set it and forget it - Community reports — neighbors confirm each other's schedules

Tech: Next.js 15, Supabase (Postgres, 954K+ records), Vercel (102kB first load), Plausible + GA4. No auth, no accounts — zero friction.

The data problem: San Diego has 500K+ households. The city's data is locked behind reCAPTCHA Enterprise on a Salesforce endpoint. My approach: build the database from the community up. Users report their pickup day → consensus verification (3+ confirmations = verified). Started with 201 addresses, expanded to 954K+ via hauler data imports across 79+ neighborhoods in 2 cities.

What I learned: 1. Address normalization is surprisingly hard (Ave vs Avenue vs Av, apartment stripping, ordinal conversion) 2. Every municipality structures trash data differently — there's no standard 3. People WANT to report their trash day. The "help your neighbors" angle converts better than any growth hack. 4. AI confidently hallucinates municipal data. This is a real gap.

Free, no ads, community-powered. Would love feedback. Planning to expand to more cities.