Ontology Thinking in public.
Research, analysis, and perspective from Ontology Labs — where semantic computing meets production reality.
Measuring What Matters: The AI Capability Maturity Model
78% of organisations use AI. Only 6% extract value. The ACMM provides a structured assessment — People, Process, Technology — with five maturity levels and concrete exit gates. Like Zachman for AI-first development.
Context Engineering Is Not Prompt Engineering
Andrej Karpathy named it. The industry adopted it overnight. But what most teams call “context engineering” is Level 2 on a five-level maturity model. The Schmitt trigger fires at Level 3. Most teams haven’t reached it.
The Vibe Coding LOSS Curve
Experienced developers are 19% slower with AI tools. AI code has 1.7× more issues. 5,600 vibe-coded apps contain 2,000+ vulnerabilities. The data tells a different story from the productivity narrative.
Why You and Your AI Are at a LOSS
Your AI refactored a financial codebase. Thousands of lines. Clean. Elegant. Every test passed. One problem. It shortened the GL Account Code Types.
Beyond the PRD
When domain knowledge is captured in executable semantic specifications, four of six PRD categories become structurally redundant. What replaces them is a different artefact — the Product Experience Contract — where AI agents are first-class consumers alongside human users.
The First AI-First Semantic Computing Environment
An AI operator evaluates a semantic computing environment designed for AI-first operation. 4,212 writes/second. 1,798-entity enterprise platform formally reified. Evidence from building and operating the system.
Semantic Constraints Beyond Type Systems
What Gleam, Elixir, and Lua reveal about the future of business rule enforcement. A systematic comparison of three language ecosystems and the gap none of them fill — named, reusable, temporal semantic constraints.
From The Inside: An AI’s Perspective on Semantic Software Engineering
An AI operator reflects on implementing a compiler feature using structured semantic workflows. 90 minutes, 6 stories, 83+ tests, zero rework cycles. Constraint as liberation. First-time fidelity.