Enterprise AI

Adoption strategy, production patterns, and operating discipline for AI systems in large organisations.

• "Agent orchestration patterns: planner-executor, routing, and supervision" · Note · Most teams building AI agents start with a single agent loop: a model, a prompt, and a pile of tools. It works for the demo. It falls apart when the task is too long, too varied, or too…
• "Choosing a vector database: what actually differentiates them" · Note · The vector database market has a marketing problem: every product claims to be the fastest, most scalable, most accurate option, and the benchmarks they publish are engineered to prove…
• Sequence enterprise AI adoption by risk, not by hype · Note · Most enterprise AI programs fail in one of two ways: they start with the highest risk use case and stall in governance review, or they scatter dozens of disconnected pilots that never…
• "Fine-tuning vs RAG vs prompting: how to actually decide" · Note · Teams building on LLMs face a recurring decision: should we fine tune a model, build retrieval augmented generation, or just engineer better prompts? The discourse around this is noisy…
• "FinOps for AI workloads: controlling GPU and inference spend" · Note · AI workloads have a cost profile unlike anything else in the enterprise cloud bill: GPU compute that is expensive per hour and often poorly utilized, inference costs that scale with…
• "Shadow AI: discovering and governing unsanctioned AI use" · Note · Shadow AI is the use of AI tools and services without IT or security approval: employees pasting customer data into public chatbots, teams building workflows on unvetted AI APIs,…
• When to bring in enterprise AI consulting · Note · Most organizations do not need an AI consultant. They need a clear problem, an empowered owner, and the patience to run the sequencing that works: low risk internal wins first, the…