I build software and AI delivery systems for mature product teams.
For more than 20 years I have worked close to the code: founder, IC, architect, product builder, and engineering leader. Today I am turning AI coding from individual experimentation into repeatable team leverage.
AI helps me move faster, but the useful part is still judgment: reading the code, understanding the tradeoffs, keeping the quality bar high, and knowing when an agent is drifting.
- AI delivery systems: skills, rules, review loops, agent instructions, background-agent workflows, and verification gates.
- Developer tools for local workflow, parallel work, review, CI, and agent-driven development.
- Product software in mature codebases, often Rails, with real customers and production constraints.
I am working on how experienced developers can use AI coding tools without giving up engineering ownership.
That means building the machinery around the tools: scoped workflows, reusable prompts and skills, review agents, local verification, background-agent experiments, and better ways for teams to share what works.
- skills - reusable Codex/agent skills for review, writing, image generation, Rails, and workflow automation.
- stooges - copy-on-write Git workspaces for agents, hotfixes, PRs, and parallel branches.
- pi-workflows - named slash-command workflows for launching Pi subagents.
- pi-soloterm-tools - Pi tools for SoloTerm MCP: agent panes, processes, todos, and scratchpads.
- solopowers - Superpowers-style workflows adapted for SoloTerm.
- parallel_specs - an RSpec runner with a live local dashboard and balanced parallel splits.
- local_ci_plus - a Rails gem for faster local CI loops with parallel steps, fail-fast, and resume support.
- lewp - a macOS-first local domain router/proxy for parallel local development.
More projects: scottw.com/projects
- Writing: scottw.com
- Projects: scottw.com/projects
- LinkedIn: Scott Watermasysk
- X: @scottw
Built by hand. Accelerated, not automated.