Trust Doesn't Rust: Designing Auditable AI Agent SDLC Pipeline

Most AI-assisted development workflows have a black box problem — agents generate code, specs, and decisions, but the artifact trail evaporates. When something breaks (or a stakeholder asks "why did it do that?"), there's nothing to point to.

This session walks through a spec-driven, agentic SDLC we built and battle-tested in a live community build cohort (Atlanta Tech Community Network, Cohort #1). Every stage — PRD → technical spec → implementation → QA → DevOps — is gated with a "works cited" checkpoint: structured artifacts that show what the agents produced, what sources they used, and what decisions were made before the next stage opened.

You'll leave with:
- A blueprint for a multi-agent pipeline with clear handoff gates
- The "works cited" pattern for AI artifact provenance
- Lessons from running this as a teaching vehicle with real contributors and a real product shipping real code

(This is the system that I built for my current job (Connexure) and I teach as part of Atlanta Tech Community Network)

Don't Outsource Judgment: Staying Human in an AI-First World

For the first time in our careers, we're working alongside systems that sound more confident than we feel. AI doesn't hesitate. It doesn't hedge. It produces answers with authority — even when those answers are wrong. And that changes how engineers think, decide, and ship.

This session explores the psychological and architectural impact of that shift: how AI-driven development creates a new kind of imposter syndrome, how over-trust cascades into silent failures, and what engineering teams can do about it.

Through real-world patterns drawn from half a decade designing LLMs and leading AI-forward engineering teams, attendees will leave with concrete tools for staying in the judgment seat — including workflow designs that require verification, a team protocol for challenging AI output, and a reframe on what "human-in-the-loop" actually demands.

Keynote presentation: Nashville AI Week, May 2026

Takeaways: AI fluency is not AI deference. Judgment is infrastructure. Confidence was never the signal — and it's time we designed like we know that.

Audience: Engineers, engineering leads, and technical practitioners working with or building on AI tools. No ML background required.

Format: Keynote, 45–60 min.

In Pursuit of Memory: Building AI Systems That Don't Forget

Every AI system has the same silent failure mode: it forgets. The model is brilliant. The session dies. The context window closes. The decisions, the corrections, the "we already tried that" — gone.

This talk is about solving that. Drew Schillinger, Enterprise Architect at Connexure, walks through the full memory stack he's built on top of Claude Code and Anthropic's model: a spec-driven workflow (gstack / drew stack), a cross-session issue tracker that doubles as institutional memory (beads), a pgvector database Claude can query at inference time (ghost. Build), hooks that enforce discipline at every tool call, and Rick and Morty. Lots of Rick and Morty.

Along the way: Claude's 5 practitioner domains, a frank look at the model spec and what happens when you test its edges, and a live build adding a new agent to a running production app — spec to deploy, on stage.

You'll leave with a mental model for AI memory infrastructure, a working understanding of the drew stack workflow, and a concrete pattern for building systems that get smarter across sessions

Nashville AI Week, May 2026

From Prompt to Product: Agentic Workflows in Practice

In this live build session, I walked through taking an idea from prompt to production using agentic AI workflows.

The talk focused on how to move beyond prompt experimentation into structured, spec-driven systems: defining clear contracts, orchestrating subagents, introducing human-in-the-loop gates, and verifying outputs before they reach users.

Using a real-world Atlanta-focused application (a FIFA/MARTA traffic navigator), I demonstrated how these patterns translate into working software—bridging the gap between AI prototypes and production-ready systems.

This session was presented at both Atlanta AI Week and GitHub Copilot DevDays at Microsoft.

Atlanta AI Week, April 2026

Github Dev Days, April 2026