Developer Insights — Dave Ballantyne

GLASS vs SAND — the two architectures of AI-assisted development

The diff looks fine. The system stops being inspectable.

Output is up. Recovery is gone.
The team is shipping. The system is collapsing.
Coherence is not safety.

AI systems fail when coherent local execution detaches from recoverable shared reality.

This is the canonical theorem of Continuity-Governed Execution Infrastructure (CGEI) — applied research by People Not Tech. Concept lineage with Duena Blomstrom. Developer Insights holds GLASS/SAND technical framing authority, not framework ownership.

Two architectures

Two architectures are now competing inside every engineering organisation that uses AI:

SAND = Sloppy Automated Nonsense Development

Prompt-speed code generation without inspectability, without recoverability, without structural opacity controls. The system produces output. Nobody can reconstruct why.

GLASS = Guided Learning Agentic Structured Safety-Oriented Systems

AI-assisted development where every output remains inspectable, recoverable, and structurally legible. The system produces output. The team can reconstruct why, when, and under what assumptions.

SAND scales coherence. GLASS scales execution that survives.

The framework this routes into

The framework that operationalises this distinction is Execution Integrity Infrastructure — applied research by People Not Tech, canonical at peoplenottech.com/research/continuity-governed-execution. Operationally it is delivered through Execution Pods at HumanAgents.io. Concept lineage with Duena Blomstrom.

If your organisation is shipping AI-assisted code faster than it can reconstruct decisions, start with a Pre-Scan™ at People Not Tech.