Brad Spires

I explore what AI makes possible and build technology for consequential work.

Brad Spires

Professional

I've spent most of my 29-year career as a hands-on technology leader, consulting with stakeholders and building business-critical systems-of-record in varied industries: government operations, financial accounting, e-commerce, and consumer lending. From "ride or die" startups to huge multi-national financial institutions, I've successfully stewarded the development of dozens of core business systems from 0 to 1. I've been refining my agentic coding process daily since 2024, and I can leverage it as a significant force-multiplier. I have worked 100% remotely for over 20 years.

Tech Leadership

  • Head of Applied AI: Built AI-native systems full time from 2022 to 2026 for Courthouse Computer Systems (CCS), with four successful deployments.
  • #1 software vendor for NC RODs: As tech lead, helped Courthouse Computer Systems (CCS) grow from a two-person startup in 2004 into a North Carolina success story. In 2026, CCS is a market leader with their system-of-record deployed in dozens of NC counties, customers in five states, and operations on four continents.
  • #1 NC Startup Exit: Led the vendor tools team building cross-market listing technology for AuctionRover.com. The $169 million acquisition was the largest VC-backed acquisition of the year in NC.
  • Award-winning tech builder at a Top 5 Bank: Designed and built a meta search system for siloed accounting data stores at First Union Bank (now Wells Fargo Bank). The project was awarded the company's President's Award for Innovation.

Skills

  • Consulting with stakeholders to identify automation opportunities.
  • Owning and building full-stack applied AI systems, from 0 to 1.
  • Modernizing legacy software with frontier AI models and workflow automation.
  • Integrating greenfield and legacy products with third-party systems.
  • Fluent in C-suite and geek speak.
  • Working within the constraints of highly regulated organizations.
  • Began developing integrated ML, GenAI, and Reinforcement Learning systems starting in 2015.
  • Agent-first coding since 2024.
  • Python/TypeScript/C#, PostgreSQL/RDS/Dynamo, Codex/Claude, OpenAI/Anthropic/Hugging Face/Bedrock, Cost Engineering/Supervised Autonomous Agents/Document Workflows/Voice Assistants/Offline Authentication/Semantic Search, Android/iOS/Linux/Windows.

Projects (authentication uses AWS Cognito: Demo username and password)

Check out some of my recent projects, prototypes, and demos.

Document Automation

  • K-Doc Vision — Extract information from document images into structured data.
  • K-Sem Search — Turn your documents into knowledge you can search, question, and verify. Includes a basic semantic search "ingestion to question" stack.

Digital Assistants

  • K-Voice Assistant — Build voice assistants using reusable building blocks. Includes a freight load booking prototype implementation.

Notes From The Frontier

After 25+ years of hands-on coding, 2 years of managing coding agents 40 hours per week, and successfully deploying agent-backed workflows to multiple organizations, I've gained a certain perspective.

Agentic Coding

  1. Primacy Of The Spec
    The single most important step in agentic coding is creating the design spec. When you take the time to get it right, the build process can be a breeze.
  2. KYT (Know Your Team)
    I had to internalize that the coding agent and I are a team, each with strengths and weaknesses. In my experience, the coding agent almost always knows more about what is possible than me, but I know far better than the agent what my stakeholders need. Don't let what you think you want or need prevent you from exploring what is possible.
  3. Force Multipliers
    When managed and guided properly, frontier coding agents are a significant and lasting force multiplier.
  4. Two Direct Reports
    For projects that matter, I usually manage no more than two agents simultaneously. Beyond that, my ability to monitor and direct each agent's progress is degraded.

Automation In The Office

  1. Gain Staff Trust
    If your automation goal includes staff-facing tools or workflows, you have to have a minimum level of staff buy-in to accurately assess the automation's success and avoid "reputational sabotage" of the automation.
  2. Labor Competition
    Anthropomorphizing an AI-backed system by calling it "he" or "she" or "a member of the team" is a bad idea and a sure-fire way to get staff pushback.
  3. Assistants, Not Replacements
    To improve employee buy-in, focus your message on how the AI-backed system supports employees' success on performance review metrics. For example, "accuracy" or "task completions per hour".

After Work

I'm a proud father and husband, a recovering singer-songwriter, and an advocate for childhood retinoblastoma awareness.

In my free time, I explore questions around navigational network encoding in prehistoric earthworks of North and South America. I study evidence of seasonal and long-term migration corridors using satellite imagery, vision models, custom lidar tools, and a purpose-built multi-modal semantic repository (photos, gis, documents).

Two adults and a child in costumes outdoors
Four aerial and terrain images of earthworks and landscape