Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
Building a System That Knows Why It Changed
This talk shows a system that learns from failures by committing to expected outcomes and using structured debug reports, making self-improvement verifiable and repeatable.
consulting-os is a git-versioned personal OS for AI agents: plain files that give Claude typed memory, composable role-separated skills, and a learning loop that persists across sessions.
For the demo, I walk through a real artifact chain. A plan written before any code ran. A debug report where an agent names its own shortcuts, addressed to a separate role for triage. The regression that diagnosed the first fix as wrong-shaped. The meta-heuristic that generalized the lesson. The chain closes with the system catching its own gap while building this talk and shipping a structural fix the same day.
Defines structured schemas using JSONL, YAML, and Markdown files.
- claude-sonnet-4-6Claude 3.5 Sonnet sets the industry standard for intelligence and speed, outperforming competitors on key coding and reasoning benchmarks.Anthropic’s mid-tier powerhouse, Claude 3.5 Sonnet, delivers frontier-level performance at a fraction of the cost. It operates at twice the speed of Claude 3 Opus and consistently beats GPT-4o in graduate-level reasoning (GPQA) and coding proficiency (HumanEval). With a 200k-token context window and the new Artifacts workspace, it transforms raw data into functional code, vector graphics, and UI components in real time. This model is the definitive choice for developers building complex, high-throughput applications that require nuanced instruction following and rapid execution.
- claude-opus-4-7Claude 3 Opus is Anthropic’s most capable model, outperforming GPT-4 on industry benchmarks like MMLU and HumanEval.Claude 3 Opus serves as the flagship model for high-complexity tasks, delivering near-human levels of comprehension and fluency. It features a 200,000-token context window (expandable to 1 million for specific use cases) and excels at open-ended questions, needle-in-a-haystack retrieval, and complex financial analysis. By prioritizing safety and steering clear of harmful outputs, Opus provides a reliable foundation for enterprise-grade automation and sophisticated research.
- VS CodeVS Code is the free, open-source code editor from Microsoft: cross-platform, fast, and feature-rich for modern development.Visual Studio Code (VS Code) delivers a powerful, lightweight development environment for Windows, macOS, and Linux. It’s built on open source and features essential tools right out of the box: integrated Git control, robust debugging capabilities, and intelligent code completion (IntelliSense). The platform’s strength lies in its vast Extension Marketplace, allowing developers to quickly add support for virtually any language or framework, from Python and JavaScript to C++ and Docker. This editor is the definitive choice for millions, balancing simplicity with enterprise-grade functionality.
Compose Email
Loading recent emails...