Meta
Starting Bourne Forge AI
Why this site exists: a place to document real AI experiments, technical observations and practical lessons rather than polished marketing copy.
Notes
Practical articles and technical observations drawn from real AI experiments, systems engineering, and more than four decades of Internet architecture experience. New notes are published as projects produce lessons worth documenting.
Internet HistoryInfrastructureArchitecture
From ARPANET to the Internet - a personal journey
How packet switching, ARPANET and TCP/IP became the architecture of the modern Internet, including a firsthand University of Tasmania perspective on ACSNet and AARNet. Three parts.
Read the full note →Prompt and Workflow Experiments: What Actually Improves AI Results?
Why the quality of AI-assisted work has less to do with finding the perfect prompt and more to do with the workflow, evaluation framework and systems architecture built around it. Eight chapters, best read in order.
AI Infrastructure & Architecture Series
A ten-part series applying decades of Internet architecture experience to building reliable, production-grade AI systems — best read in order. Starting July 13, 2026, a new part publishes every two weeks.
Stay tuned
Part 6 of 10, Model Independence and the Portability Layer, publishes September 21, 2026.
Part 1 of 10ReliabilityAI
Defining Reliability for AI Systems
Why uptime is not enough when outputs can be available, fluent and wrong: a measurable framework covering SLIs, error budgets and acceptable failure for AI workflows.
Read the full note →Part 2 of 10ReliabilityAI
Designing AI Systems for Failure
Retries, fallbacks, circuit breakers and graceful degradation for model-dependent applications, plus a failure-mode matrix for mapping failures to responses in advance.
Read the full note →Part 3 of 10ObservabilityAI
Observability for Non-Deterministic Systems
How to understand an AI application when identical inputs do not guarantee identical outputs: versioning, tracing, retrieval quality, and catching gradual quality drift before users do.
Read the full note →Part 4 of 10ContextAI
Context Architecture: The System Behind the Prompt
Why information provenance, freshness and retrieval matter more than clever prompting: source authority, chunking, ranking, context poisoning and showing citations to users.
Read the full note →Part 5 of 10SecurityAI
Securing Tool-Using AI Systems
Prompt injection, excessive agency and the dangers of connecting language models to real infrastructure: least-privilege permissions, approval gates, credential isolation and a tested kill switch.
Read the full note →Other Notes
InfrastructureAI
Internet Architecture Lessons for AI Systems
What 40 years of building the Internet can teach us about building reliable AI: layered architecture, graceful degradation, loose coupling, observability and more.
Read the full note →Evaluation
Evaluating Practical AI Tools
Looking beyond the hype to find AI that actually delivers: a practical framework covering reliability, total cost, integration, transparency and failure behaviour.
Read the full note →That's everything so far — new notes are added as experiments produce something worth writing down. Get in touch if there's a topic worth covering.
Published