SERIES

Enterprise AI

Ten lessons on making a model into a system. Where it runs, who approves an irreversible step, and what a run costs.

This series is not legal advice. The roadmap lesson uses planning bands, not a Cloudadorn delivery quote.

  1. 01 A model is not a system A model drafts and scores text. A system adds context, tools, checks, and an owner. Fluent output is not correctness, permission, or a finished job. 3 min read
  2. 02 Local, sovereign, and hybrid Local means the model and the harness run on hardware you control. Sovereign means the boundary holds. Hybrid routes each job by the data and the risk. 3 min read
  3. 03 The loop inside the harness The harness is context, state, tools, a check, permission, and a trace. The loop reasons, acts, and looks. The model is one part of that loop. 3 min read
  4. 04 Context, and when it goes stale The window is a budget. Longer input can get less reliable. Retrieve the evidence for this step, and test that choice on your own workload. 3 min read
  5. 05 A human at the irreversible step The agent may prepare the action. A person approves the step you cannot cheaply undo. Review is a control, and tired review stops being one. 3 min read
  6. 06 Evals and release gates A golden set is your real cases, run again after every meaningful change. A model judge has to be checked against people. Public benchmarks are not your gate. 3 min read
  7. 07 What a run actually costs Count the cost of an accepted outcome, not the cost of a token. Batch and cache change the hosted bill. Spare capacity is not free headroom. 3 min read
  8. 08 Six things a company can build Documents, search, meetings, a private model, a multi-step agent, or a capability designed backward from the outcome. Pick one. Do not boil the company. 3 min read
  9. 09 A reference picture and a first roadmap Draw where a request goes, who approves it, and how you will know. Then run one bounded pilot. Week ranges in a method are not a delivery quote. 4 min read
  10. 10 A weekly scorecard Once a week, look at quality, retrieval, the path the agent took, security, speed, cost per accepted outcome, and whether people use it. 3 min read

Talk with an engineer