Justin Heath

I design how engineers and AI agents build software together.

Twenty-five years in .NET and data platforms, usually where the process is broken, manual, or both. Now I work on AI enablement and agentic systems: the context agents work from, the specifications and verification gates their work has to pass, and the reviews that check their output against code and data.

What I'm Working On

workit

An open-source Claude Code plugin with the engineering workflows I use and maintain: specifications with review gates, work-package execution, cross-vendor code review, diagnosis, and session handoffs. Skills load detail only when a step needs it, and scripts run the steps that shouldn't vary, such as starting agent sessions and structuring review output.
Observatory

My personal multi-agent platform for engineering and research. Memory and working context carry across chat, a web workspace, and the CLI, backed by PostgreSQL and pgvector. It runs scheduled research briefings and cross-vendor review that confirms or refutes each finding. The code is private, and I'm glad to demo it. One account of it in use
Ideas — Cognitive Architecture & Agentic Workflows

Essays on designing AI systems around how people actually think. Cognitive fit, specification literacy, agent architecture, and lessons from building an auto-healing dispatch system.
The Discovery Journal

A guided self-examination tool for discovering your cognitive architecture. Interactive web app, browser-native, no backend.
Family Kitchen

Meal planning to recipes to shopping lists, built for my actual household. It started as a Blazor Server app; now it's a Svelte 5 front end on a .NET 10 API with PostgreSQL, running in Docker. GitHub
Schedule I Mod

MelonLoader mod for Schedule I with C# support for both Mono and IL2CPP runtimes. GitHub

Writing / Thinking

Short versions here. Longer essays at ideas.heathdev.me.

Friction as Telemetry

Most productivity advice assumes friction is a character flaw. After twenty years of failed systems, I stopped optimizing the system and profiled the operator instead. The real tax wasn't inefficiency — it was misalignment. Friction isn't always resistance. Sometimes it's telemetry.

Specification Is the New Literacy

When a first implementation takes minutes, the bottleneck moves to knowing what to build. AI fills ambiguity with plausible-looking software that compiles and has nothing to do with what the customer needs. The skill that matters now is describing what you want precisely enough to build it and check it.

Read the full essay →

How Your Mind Actually Works

Most design asks "what do you want?" — the wrong question. The right question is "how does your mind actually move?" I treated my own brain like a poorly-documented legacy system and ran a structured analysis. The highest-leverage act in building personal software is discovering how the person using it actually thinks.

Read the full essay →

You Don't Trust — You Instrument

Trust isn't the mechanism for agent-written code. You define scenarios outside the codebase, let the agent build, then check its behavior against them. The critical principle: the thing being validated cannot also control the validation criteria. Scenarios add to code review rather than replace it; someone still reads the architecture, the security boundaries, and the code.

The Expertise Paradox

Deep implementation expertise can be a liability in AI-augmented work. Generalists often adopt AI faster because their native workflow is already "I know what this should do, I'll look up the how." The expert's identity is forged around implementation — redirecting it to specification and evaluation is the necessary transition.

Read the full essay →

Externalized Intent

If it only exists in the code, it doesn't exist. Code captures what and how, but intent — the why — evaporates unless deliberately externalized. Documentation, tests, and visualization each serve a different cognitive moment. Without all three, every pause becomes a dig site. With them, every pause is resumable.

Background

Started in a hospital basement in 2000, routing calls through a switchboard database. Twenty-five years later, I still end up in the same kind of work: systems that should talk to each other but don’t, and processes running on patience instead of automation.
Eight years at Horizon Hobby, from support to building their B2B ordering platform and bridging .NET systems to an IBM AS/400 and, later, a System i5. That’s where I learned the interesting problems are usually at the seams.
Thirteen years at Volition, the studio behind Saints Row and Red Faction. I led IT development across HRIS, payroll automation, SharePoint migrations, and internal tools, and cut HR payroll processing time in half. The studio closed in August 2023.
Now a senior software engineer at Veson Nautical, working across Q88’s .NET and SQL Server applications and the Veson Platform’s Node.js stack. I was the main engineer on Q88’s SIRE 2.0 inspection module, from PDF ingestion through fleet analytics. I also designed and built both sides of Q88’s integration into the Veson Platform. On the AI side, I built the shared context library the team’s AI configuration imports and published two AI-engineering guides in Veson’s engineering knowledge base. I run multi-agent code reviews, with findings checked against code and data.
The constant: find the manual, fragile thing held together by institutional knowledge, then replace the fragility with a system that lasts.