Agent frameworks
Running, v2 in progress
Tommy: a 20-agent OpenClaw framework that runs my day, my research and my health plan
It started as one assistant in a Telegram thread. It is now a framework of twenty agents on OpenClaw, each with its own job, its own rules and its own workspace, sharing one brain and eleven connectors. Tommy is the one I talk to. The rest report in.
- Result
- Twenty agents in five teams, one shared brain, and a front door I can reach from my phone.
- Roadmap
- In progress: Tommy 2.0. Open-weight models in the cloud for the routine work, frontier models only where judgment matters, spend caps per agent, and a watchdog that restarts anything that falls over.
Read the build →
Multi-agent workflows
Running
Mission control: the router, the shared brain and the runbook behind dozens of AI jobs
Every morning dozens of scheduled AI jobs wake up across the cloud and a Mac mini: ad monitors, content runs, research desks, store upkeep. Before any of them does real work it asks a router which model to use, reads the same brain as everyone else, and follows one runbook. That is the bit nobody posts about, and it is the bit that makes the rest work.
- Result
- One router, one brain and one rulebook shared by every agent and every scheduled job.
- Roadmap
- Next: per-job cost reporting from the router log, and automatic retirement of jobs that have not changed a decision in a month.
Read the build →
Investor desk
Running
Gordon's desk: a 14-agent research team with a risk gate that can veto my own ideas
Gordon used to write one letter on a Sunday. Now he runs a desk: thirteen specialist agents covering news, earnings, macro, charts, value and risk, a red team paid to disagree, thesis cards with kill criteria, and a risk gate that blocks any proposal that breaks the rules I set. It never trades. It argues.
- Result
- A daily advisor note and a Sunday deep dive that grade their own past calls.
- Roadmap
- Next: a longer track record before any rule is trusted, and the desks' hit rates feeding how much weight each one gets.
Personal research tool. Not financial advice, not a recommendation to buy or sell anything.
Read the build →
Trading bots
Paper trading
Two trading engines where code holds the risk and AI only picks from a menu
A crypto engine on OKX and a second bot that trades crypto on Binance and US stocks on Tiger Brokers. Code generates the trades, sizes them and manages every exit. A fast model picks one option from the menu or says hold. A slow model sets the mood once every few hours. Neither is allowed near the stop-loss.
- Result
- Two engines running on paper with every decision, fee and model call logged, and a promotion gate that has already said no once.
- Roadmap
- Next: fix the repeated-signal bug and the slow layer's unreadable replies, then a fresh 14-day run on the exchange's demo engine. Live keys only if it clears the gate.
Personal experiment. Not financial advice, not a recommendation to buy or sell anything.
Read the build →
E-commerce
Running
Tail Friends: a pet store that AI built in a week and now runs every day
A real Shopify brand, live in 18 countries. AI researched the products, wrote every listing, set the prices, made the videos and the music, runs the social accounts, watches the ads twice a day and pulls the plug on losers, finds new products every Sunday and hands me a KPI review every Monday. I approve. It does the rest.
- Result
- Around 40 products, 3 social channels, paid ads with automatic kill rules, and the first orders in, run by agents with one person approving.
- Roadmap
- Next: reviews on the best sellers, growing the creator program (commission paid only on products with a healthy margin), and ads back on only behind written caps.
Read the build →
Content and socials
Running
A content engine for two brands: research, images, video, social packs and one review sheet
Every week a scheduled job researches the stories worth covering, checks each claim against at least five recent sources, makes the hero images and charts, and cuts every piece into a Medium copy, an X post, a LinkedIn version and an Instagram square. The store gets its own version for short video. Everything waits in one review sheet. Nothing ships without my yes.
- Result
- A full weekly content calendar for two brands with a ten-minute daily review as the only human step.
- Roadmap
- Next: The Build Log newsletter sent from the same pipeline, and short video for this site, not just the store.
Read the build →
Personal systems
Running
Abby: a health coach agent that reads my bloodwork and checks in twice a day
Every lab report, blood panel and wearable export goes into one place. Abby, a coaching agent with her own goals file, diet plan and honesty rules, reads them, tracks what moved, keeps the plan current and checks in every morning and evening. She is polite about it. She is also relentless.
- Result
- One structured record of everything, and a daily nudge that knows what I did yesterday.
- Roadmap
- Next: trend charts per marker across years, a doctor-ready summary before every appointment, and grocery lists generated from the nutrition plan.
Read the build →
Local-first infrastructure
Running
A smart home system with no cloud, no subscription, no vendor
I replaced a commercial home controller with a system I built myself. It runs on my own hardware, nothing it does leaves the house, and I can ask Tommy to turn the lights off from the other side of the world.
- Result
- Zero cloud calls, zero subscriptions, zero monthly cost.
- Roadmap
- Next: energy monitoring per circuit, presence-based automation, and the full device fit-out for the new apartment (see Marcel).
Read the build →
Project agents
Running
Marcel: the renovation designer who never forgets a decision
A full apartment renovation, two interior designers, a contractor, an electrician and a WhatsApp group with 900 messages. Marcel reads all of it and keeps the single list of what was actually agreed, room by room.
- Result
- One dated, room-by-room master list where the latest instruction always wins.
- Roadmap
- Next: photo check-ins from site against the plan, and an automatic snag list before handover.
Read the build →
Sales and business systems
In use
A GTM engine: account research, daily signal scans and a memo in my inbox before 8am
Give it a target list. It enriches every account, scores them on timing rather than size, watches their websites and job boards every day, writes the morning memo, and drafts the outreach that reps actually send.
- Result
- A full account playbook from a company name and a page of notes, refreshed daily.
- Roadmap
- Next: reply routing into the CRM, and reporting split by segment and persona so the aggregate number stops hiding the answer.
Read the build →
Sales and business systems
In use
A messaging engine that turns customer conversations into what reps should say next
A knowledge layer on a company's products, services and competitors, fed with call transcripts, customer emails and support tickets. Out the other side come value propositions, opportunity ideas and customer goals, written as clear messaging for every rep before their next call.
- Result
- Every rep walking into calls with the customer's own words, goals and gaps in front of them.
- Roadmap
- Next: a per-account brief that refreshes the night before every meeting, and a scored list of expansion ideas ranked by evidence.
Read the build →
Personal systems
Running
One place that knows every policy I hold, and what it does not cover
Life, health, home, travel, an investment-linked plan, and the fine print on all of them. An agent reads the documents, keeps a coverage map, flags gaps and overlaps, and reminds me before anything renews.
- Result
- A coverage map I can answer questions from in seconds instead of a weekend of PDFs.
- Roadmap
- Next: an annual review letter comparing cover to life changes, and a claims assistant that fills the forms.
Read the build →
Personal systems
Running
Ten years of iCloud, sorted by an agent that asks before it deletes
Duplicates, screenshots, downloads from 2016, three copies of every tax return. An agent walks the whole library, proposes what to merge, move or bin, and waits for a yes.
- Result
- A review queue instead of a mess, and a monthly report of what changed.
- Roadmap
- Next: automatic filing of new downloads into the taxonomy, and receipt extraction from photos.
Read the build →