Essays & a working paper
Writing
Essays on running AI as a business function: what it costs per unit of work, how to govern it before the rules arrive, and where production systems fail quietly. The engineering notes are the receipts.
Working paper
Forecast, Measure, Optimize
A workflow-level model for the value and consumption of agentic AI
Enterprise AI business cases usually produce one number, the value of deploying, and leave the second number, what it will cost to run, to be discovered in the first invoice. This paper presents a workflow-level model that produces both from the same inputs.
Economics of enterprise AI
5 essays · 32 min
Why the bill is visible and the cost is not, and how to forecast both.
- 7 min Read →
The business case for AI has two numbers. Most people only produce one.
What an AI workflow returns and what it costs to run come from the same inputs. A method for forecasting both, with a ramp, a range, and a map of when run cost decides the case.
- 5 min Read →
The unit economics problem in enterprise AI
Enterprises can see their AI bill but not their AI cost, and until spend maps to units of work, AI budgets are unforecastable.
- 6 min Read →
What is a token, really?
A token is a metered unit of computation with a physical cost floor. Budgeting for AI starts with knowing what you are buying.
- 7 min Read →
Which workflows first
Rank candidate AI workflows on two axes, value returned per dollar of run cost and governance friction, and start in the corner that is labor-heavy and low-friction. The showcase candidates are usually in the wrong corner.
- 7 min Read →
Build, buy, or partner, and the lock-in nobody budgets for
The expensive lock-in in enterprise AI is not the model but the untagged usage history, the prompt estate tuned to one vendor, and the evaluation suite that was never written. Vendor neutrality is an architecture property, and the right partner is one that helps you keep it.
Governance and accountability
4 essays · 20 min
Designing who answers for an AI system before the first workflow ships.
- 5 min Read →
Governance is the strategy, not the checkbox
Governance is the load-bearing element of AI strategy, and it has to be designed into the architecture before the first workflow ships.
- 6 min Read →
What enterprise grade actually means for AI applications
The distance between an impressive AI demo and an enterprise application is not model quality. It is identity, observability, cost attribution, failure handling, and a named owner.
- 2 min Read →
Compliance first
Auditability is a property of the architecture, not a feature added before launch. The work of making a system auditable and the work of making it correct are the same work.
- 7 min Read →
Agents need a job description
An agent in production should be governed the way a new hire is: a written scope of authority, actions it may take alone and actions that need a signature, a manager who answers for it, and a way to take it off the floor.
Running the program
3 essays · 20 min
Sequencing, platforms, and enablement: how an AI function compounds instead of restarting.
- 8 min Read →
The first ninety days of an AI program
A new AI leader should spend the first quarter on inventory, instrumentation, and one paved road, not on a flagship build. The organizations that start with a showcase spend the second year rebuilding it.
- 6 min Read →
Training is not adoption
Enterprise AI capability is built by certifying people against shipped, governed work, not by seat licenses and lunch-and-learns. Enablement is a delivery discipline with its own metrics.
- 6 min Read →
One paved road, many workflows
Enterprise AI compounds only when it is run as a platform: tenancy, a shared gateway, approved patterns, and a release gate. Run as a sequence of projects, every workflow pays the full cost of the first.
Engineering methodEngineering method
5 essays · 20 min
Building where mistakes are expensive. The receipts behind the positions above.
- 5 min Read →
Prompting is cheap. Context is expensive.
In production agentic systems, the dominant costs and failure modes come from context management, not prompt wording.
- 6 min Read →
Five tags on every model call
A five-field telemetry schema, decided before the first workflow ships, serves finance, compliance, and quality at once. It is the cheapest architectural decision an AI program will make and the most expensive to retrofit.
- 3 min Read →
Scope before you build
How tightly framing the problem decides whether an AI project ships or stalls.
- 3 min Read →
Silent failures
The most dangerous bugs in AI systems are the ones that return a confident, wrong answer.
- 3 min Read →
Exact over approximate
Why I default to exact search in production retrieval, keep it as a ground-truth oracle, and treat approximate search as a decision to justify with numbers.
The Second Number · a letter from Korey
A letter on the number most business cases leave off: what it costs to run.
One idea you can take into a board meeting, one number you can check, and nothing I would not say to a board. Written for the people who sign the budget and the people who have to deliver on it.
A few times a year, when there is something worth your time. Unsubscribe in one click. I never share the list.