AI systems / built in public

I don't write code.
I build AI systems that run without me.

Autonomous agent fleets. Local-first infrastructure. Research systems that report before I wake up. I document every build, including the parts that broke.

See the builds
  • 6 agents
  • 19 skills
  • 16 scheduled jobs
  • 8 research streams
  • 1 Mac mini

Newsletter · weekly

The Build Log

One system teardown a week. What I built, what it cost, what broke, and the exact prompts that did the heavy lifting. No tool listicles.

  • A full build teardown with architecture
  • The prompt or config that unlocked it
  • The failures, with root causes

Weekly. Unsubscribe in one click. Read by subscriber count people.

Things I've built

Real systems, running in production. Architecture, stack, and what broke, included.

Autonomous agents Running

Tommy: a six-agent fleet that runs my day

An autonomous agent system on a Mac mini that briefs me at 8am, researches on command, generates content, and ships it. I talk to it through Telegram.

  • OpenClaw
  • Telegram
  • Multi-model routing
  • LaunchAgent
  • Cron
  • +2
Result
Runs number scheduled jobs a week with no manual input.
Roadmap
In progress: credential isolation, per-agent tool policies, an exec allowlist, and a human review gate on all outbound email.
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 entirely on my own hardware and nothing it does leaves the house.

  • Python
  • FastAPI
  • Mosquitto MQTT
  • SQLite
  • YAML rules engine
  • +4
Result
number devices, zero cloud calls, zero monthly cost.
Roadmap
Next: Zigbee and Thread support, energy monitoring, presence-based automation.
Read the build
Research agents Running

Gordon: a research agent that files a weekly investment letter

Every Sunday an agent sweeps eight research streams and delivers a structured letter. It has a personality, and it does not hedge.

  • Multi-stream research
  • Brave Search
  • Structured output
  • Weekly cron
  • Telegram delivery
Result
number research streams, one letter, every Sunday.
Roadmap
Next: source citation on every claim, and a scored track record on past calls.

Personal research tool. Not financial advice, not a recommendation to buy or sell anything.

Read the build
Sales and business systems In use

AI that does the commercial work, not just the technical work

Account research, stakeholder maps, outreach sequences, and a full market-entry business plan. Same tools, applied where they pay.

  • Deep research
  • Structured frameworks
  • Market analysis
  • Outreach generation
Result
outcome, e.g. pipeline generated or hours saved
Roadmap
Next: a reusable playbook generator with a scoring model for account prioritisation.
Read the build

Take what's useful

Everything here is free. No email wall on the downloads.

  • Checklist

    The self-hosted agent security checklist

    The 12 things to lock down before you let an agent touch your email. Written after auditing my own and finding all of them open.

    Coming soon. Get notified

  • Cheat sheet

    Model routing by job type

    Which model to use for which job, and what it costs. Stop paying frontier prices for formatting tasks.

    Coming soon. Get notified

  • Blueprint

    The six-agent architecture

    How Tommy is structured: agents, skills, memory, orchestration. Copy it.

    Coming soon. Get notified

  • Prompt pack

    Prompts that force real answers

    The constraints I use to stop AI giving me hedged, agreeable rubbish.

    Coming soon. Get notified

Writing

All posts

First teardown lands soon. Subscribers get it first.

Who's building this

I’m Cian. I’m Irish, I live in Singapore, and I spent my career in commercial roles, not engineering ones. I can read code. I cannot write it well. That turned out to matter far less than I expected.

In the last year I’ve built an autonomous agent fleet, replaced a commercial home automation platform with my own local-first system, and put research agents to work on decisions I used to make on instinct. Most of it broke first. The interesting part is always why.

I write this down because the gap between what AI can do and what most people use it for is enormous, and closing it is mostly a matter of seeing someone else do it badly first, then better.

More about me

Working together

I take on a small number of build projects: agent systems, internal automation, AI where it actually pays. If you've read a teardown and thought "I need that", start with an email.

Get in touch