How we build with AI coding agents and run AI in production
50+ products built over 8+ years, and today a three-person studio running its own products on Claude Code with 30+ custom skills. These are the rules, checklists, and tools behind our 15-day MVPs and the agents we run for clients. Free to copy.
What every build follows
Think before coding
Assumptions are written down before an agent writes a line. If two readings exist, we ask.
Simplicity first
The least code that solves the problem. No speculative abstractions, no config nobody asked for.
Surgical changes
Every changed line traces to the request. Agents do not refactor what is not broken.
Verify, then ship
Every task has a check: a test, an eval, or a read-back after a write.
Human on the risky step
Sends, money, and data changes get a human checkpoint until the numbers say otherwise.
Cost per task is a feature
Every AI path has a budget, a cheaper route for simple steps, and an alert.
Engineering articles
Why We Run Every Client Project on Basecamp
One Basecamp project per client: messages, to-dos, docs, files, and schedule in one place. How we borrow Shape Up ideas and what we leave out.
EngineeringClients Get a Short Video Update, Not a Status Meeting
Why we replace status meetings with 2 to 4 minute video walkthroughs, how we record them with JiffyRecord, and the one weekly call we keep.
EngineeringLovable, Cursor, Claude, and ChatGPT: How We Use Each
Lovable for day-one prototypes, Cursor in the codebase, Claude Code for agentic builds, ChatGPT for research. When each one is the wrong tool.
EngineeringHow We Work: Build, Iterate, Grow, Run
Our four stages: a 15-day AI MVP, $1,500 weekly sprints you can pause, AI agents priced per outcome, and Agent Care. Plus what we say no to.
EngineeringThe Four Rules We Give Every AI Coding Agent
Think before coding, simplicity first, surgical changes, verified goals. The rule file behind our 15-day MVPs, with before and after examples.
EngineeringTauri v2 and Rust: The Compile Errors AI Agents Keep Making
AI agents still write Tauri v1 code and fight the Rust compiler in the same ways. The rule file we load to stop the repeat mistakes.
EngineeringStop Parsing LLM Prose: Design AI Features as Typed Decisions
Prompt, then regex the answer, is fragile. Break AI features into small typed judgments: route, select, score, verify, escalate.
EngineeringHow a Three-Person Studio Runs Seven Products on Claude Code Skills
Skills as SOPs, one task registry, research before planning, automate anything done twice, kill criteria. How a small studio stays small.
EngineeringAgents That Write to Your Tools Must Read Back
Two integration traps that cost us real work, and the rules we now enforce: replace vs patch, full objects, verify after write, plain-text contracts.
EngineeringThe Review Checklist We Run on Every AI-Generated Pull Request
Structure, error handling, performance, strict security, tests on critical paths, debuggability, and severity levels. Our checklist for AI-written code.
EngineeringCost Per Task: The One Number Every AI Feature Needs
How to measure cost per task for AI features, route simple steps to small models, cache, cap, and alert before the bill surprises you.
The tools behind every project
We pick each tool for one job and use it hard. Members of Anthropic's Claude Partner Network.
The 15-Day AI MVP Playbook
Day-by-day plan, a starter CLAUDE.md, the Lovable-to-production checklist, eval and cost worksheets, and 10 prompts. Pure value, no pitch.
I started Outcomizer because hourly billing rewards the wrong thing. As a CTO I led an 18-engineer team and shipped the products in these case studies, including a Fortune 100 shipping system and a SaaS we later sold. Now I put my fee on the result. I still do the work. You work with me, not a sales layer on top of a bench.
Book 20 minutes
Bring a URL, an app, or your lead flow. You leave with a free diagnostic and a written outcome price. The call is free either way.