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.
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- 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
- 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 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