What's been built with AI agents, and how it's going.

Reads public sources on a schedule, picks the one or two developments genuinely worth a reader's time with a source link on every claim, and says so honestly when nothing qualifies instead of publishing filler. It started in April 2026 as Scoop, a daily Alabama football brief built with an AI app builder; Scoop was paused in August and its job folded into Bama Signal.

Who it's for. Four audiences, each with its own site: medical practice leaders (Healthcare), Alabama football fans (Bama), Texas contractor finance leaders (Construction) and finance and operations leaders at smaller businesses (AI).

How it's built. One shared engine runs on Claude in the cloud, with each channel written as a plain-text playbook; it researches, drafts, gets an independent AI review, and publishes to a Next.js site on Vercel. Bama Signal adds its own football analysis: play-by-play data turned into opponent-adjusted team ratings and scouting tendencies, college football statistics, transcripts of Alabama podcasts, and a running analyst's notebook of what it has learned and how its predictions graded.

A personal health and fitness app that brings my training plan, completed workouts, recovery, food and progress into one place and turns them into a practical next step for today.

Who it's for. Me: it was built for a market of one. A public demo with a fictional client shows how it thinks.

How it's built. A Next.js web app on Vercel with a Supabase database, built by directing AI agents (first a Product and a CTO agent, now Claude Code); Claude writes the daily brief, answers questions and reads food photos, and workouts and recovery data flow in from wearables.

Keeps projects, decisions and working rules in plain files John owns, so any AI assistant can pick up the work without him repeating himself. Around it sits a small team of AI agents with defined jobs: a Product agent and a CTO agent that hand work to each other in writing, an executive-assistant agent for personal admin, and scheduled jobs that publish the Signals and check on the other projects.

Who it's for. Me and whichever AI I work with. It's also the model for the free 'Jarvis Starter Kit' I plan to offer AI Honestly readers.

How it's built. An Obsidian vault of plain Markdown files (one current-state page and one log per project, short decision records), with small adapter files that point Claude, Codex or any future tool at the same instructions. Every agent works under the same three-level rule, tested in an August 2026 pilot: proceed, proceed and report, or stop and ask. Anything consequential, like spending money, publishing or sending in John's name, waits for his approval.

Keeps my household books with a CPA-style monthly close: upload the bank and card statements, code transactions with AI help, reconcile every account, review the financial statements, then freeze the month.

Who it's for. My household. I run the close myself.

How it's built. A Next.js web app on Vercel with a Supabase database, built with AI agents; an AI coding pass follows a household rulebook I own, and the reconciliation and review steps are there so I can defend the result.

A phone-first household app for the grocery list, shopping trips, receipts and price history, recipes and meal planning, a shared restaurant list, shared to-dos and a shared calendar.

Who it's for. My household. It started as a way to stop the 'which tortillas?' texts from the grocery store.

How it's built. Built with AI agents on ChatGPT Sites (OpenAI's app hosting) with a small database; AI reads receipts and recipe photos into editable drafts, and recipes come from the TheMealDB database.

Follow podcasts, get short free summaries of new episodes, and ask for a full brief, with a short spoken audio version, when an episode looks worth it.

Who it's for. Podcast listeners who want the gist before committing an hour.

How it's built. A web app on Vercel with a Supabase database; it transcribes episodes, has Claude summarize them and turns the summary into audio.

A weekly LinkedIn series about learning AI in public as a businessperson: what worked, what broke and what I changed my mind about, with the AI getting the last word in every post.

Who it's for. Business leaders and professionals who want practical, candid takes rather than engineer-speak or recycled hype.

How it's built. AI agents research possible topics and interview me one question at a time, then draft; I write the headline, edit the draft and give final approval, and every post says what AI actually did.