The lab publishes what its systems do and what they measure, not how they are built. Results appear here as they are earned.
Software Stability Index (SSI)
Measuring what breaks, and what it costs to stand still
The Software Stability Index is a longitudinal measure of how stable software technologies remain in real use: how often they break, what it costs to keep up, how well documentation and community answers age. Scores are earned from recorded observations, never asserted, and none are published before the data supports them. The same methodology is being extended to track LLM model performance over time. The reasoning is laid out in the launch article and in a 16-minute talk on the hidden cost of over-rushed innovation.
Auditable AI code generation
Code you can review, not a black box
The lab's generators pair deterministic templates with LLM assistance, and generated output is audited before it is accepted. The claim: code you can review the way you review a colleague's pull request. Reliability and reviewability come first; generation speed is a consequence, not the goal.
Process-first AI orchestration
Deterministic by default, agentic when useful
A business-process and rule-engine platform designed to drive AI agents. With no model present, the same processes run as fully deterministic workflows, so nothing critical depends on an LLM being available. It already runs the lab's internal operations.
LLM-integrated native-app generation
SwiftUI and SwiftData, unreleased
SwiftUI and SwiftData generators exist in the lab and remain unreleased. They ship when their built-in LLM integration is complete.
Email-authentication threat intelligence
From DMARC reports to named IP addresses
Aggregated DMARC reports name the IP addresses that impersonate a domain. DMARC Aide ships that analysis today on iOS and macOS; current work extends it toward an AI-assisted web application firewall.
Local LLMs in PHP web development
Ordinary hardware, useful results
What locally hosted models can automate reliably inside PHP web applications on ordinary hardware, and where deterministic tooling still wins. Findings feed the lab's generators and its tutorials.