Tyson Hummel

Software & Systems Architect

I design systems - and the systems that build them.

20+ years building production software for banks, retailers, and insurers. Creator of DinoStack, an open-source methodology for operating AI coding agents across nine tools.

What I work on

Less time writing code by hand. More time specifying and verifying.

As AI gets good enough to write and finish code accurately, the scarce resource stops being whether the machine can build something and becomes where the engineer points their attention.

The job shifts. That shift is what I work on.

Selected work

DinoStack

An open-source methodology for operating AI coding agents. It concentrates the operator's attention at two gates, specification and verification, and runs the work in between under a structure that keeps output thorough and trustworthy.

Apache-2.09 tool adapters16 specialist agents18 workflow commands
Role
Creator & maintainer
Started
2025
Status
Actively developed
Docs
docs.dinostack.ai

Operator

Specify

Say what to build and what done looks like.

DinoStack

Build

DinoStack does the building and holds the work to your spec.

Operator

Verify

Review the output and tweak the project or task specs as needed.

DinoStack: Agentic Engineering Framework overview. Seven capabilities - classifies risk, delegates to specialists, reviews adversarially, verifies behavior, manages context, self-improving, and portable across Claude Code, Cursor, Codex, Gemini, and more.
The methodology runs across nine AI coding tools, from Claude Code to Cursor to Codex.

Independent Review

An adversarial review loop where the agent that writes the code never reviews its own work.

Built For Scale

One methodology with 9 tool adapters, 16 named specialist agents, and 18 workflow commands.

Open Source

Apache-2.0 licensed and actively developed, with the full methodology documented at docs.dinostack.ai.

Read the DinoStack docs

Open-sourced through Space Dinosaurs. Unleashed From Extinction - Rewired For Retail Innovation

SignalBrief

An AI-assisted workspace for accelerating learning. Give it a single source - a link, an article, or a video - and SignalBrief gathers more information on the topic and distills it into one document with a companion slide deck.

Learning acceleratorWorkspace templateMarp slidesClaude CodeCodex
Role
Creator & maintainer
Status
Open source
Repo
github.com/fullmetalblanket/signal-brief
Format
Workspace template

SignalBrief

One source, one document

Give it a single link and SignalBrief gathers the good information about the topic, then distills everything into one document you can keep.

SignalBrief

A companion deck

A bundled Marp workflow turns the document into a concise slide deck, so the key points are ready to present or review in minutes.

SignalBrief

Portable by design

Canonical instructions live in skills/, with lightweight adapters for Claude Code and Codex instead of duplicated implementation.

View the SignalBrief repo

Capabilities

What I've spent two decades building

  • AI-Powered Automation
  • Intelligent Assistants & Chatbots
  • Content Generation Systems
  • AI Strategy & Implementation
  • Custom AI Tool Development
  • AI Integration into Existing Systems
  • Business Process Automation
  • Automated Reporting & Alerts
  • Data Collection & Processing
  • System Integrations
  • Scheduled Task Automation
  • No-Touch Content Pipelines
  • Custom Business Applications
  • E-Commerce Solutions
  • User-Friendly Interfaces
  • Mobile-Responsive Design
  • Database Design & Management
  • Secure Payment Processing
  • Technical Team Leadership
  • Project Planning & Execution
  • Client Communication
  • Technology Roadmapping
  • Mentoring & Training
  • Cross-Functional Collaboration

Contact

Get in touch

Working on DinoStack or agentic engineering? Let's talk.