About CivicWork

Dustin Good

I'm a city council member in Elgin, Illinois serving my second term, with four and a half years of elected experience. I've voted on emergency aid for homeless residents after an encampment fire, navigated how local government protects people from opaque and overly aggressive federal immigration practices, and pushed for accountability when city-funded organizations present gaps in service.

These aren't abstract policy debates. They're non-verifiable decisions. There's no algorithm for weighing emergency shelter costs against long-term housing strategy. No optimization function for balancing federal relations against protecting our neighbors. No correct answer for how much underperformance justifies contract revisions. These are value-laden tradeoffs requiring human judgment and democratic accountability.

That's exactly why I build AI tools for this context. Not to make these decisions, but to improve the quality of research, reasoning, and collaboration that informs them.

Practitioner who builds.

My open-source work has influenced product development at Coda. The Coda MCP server, a 34-tool integration layer for AI systems, has been used by Coda's engineering team to inform their official implementation. As Coda engineer Bharat Batra noted: “We've been playing with your MCP as well and it has helped us think through our own design.”

I created CivicAide, featuring PolicyAide, an 8-agent system for policy research that uses tournament-style competition between specialized AI agents to stress-test policy approaches. Drawing on Google's AI CoScientist research and Anthropic's work on debate-based alignment, proposals compete head-to-head, weak arguments get eliminated, and stronger positions emerge through adversarial iteration.

This work led me to develop the Verifiability Framework and the Trust Stack, companion diagnostics for determining when AI should automate government processes versus augment human judgment, and what trust infrastructure strengthens different types of deployment.

I'm an AI-forward developer. I build by directing AI systems rather than writing code in the traditional sense. My preferred tools come from Anthropic; I appreciate both the capability of their models and their commitment to AI safety research. But CivicAide is architected to be model-agnostic; the governance frameworks and agent pipelines don't depend on any single provider. Government procurement requires that kind of independence, and the architecture reflects it. I work as a Coda developer for Western Governors University. Before that, I spent years in startup environments, which gave me a bias toward accountability, clear metrics, and shipping things that work.