AI & connected workflows
Intelligence needs a clear path.
Useful automation connects data, decisions and people. Define the workflow before choosing how much autonomy it needs.
Explore the approach ↗UgurKeser.art
AI and automation
Connecting technology to concrete business problems.
Methods & practical notes
AI & automation · Article
Cost and latency budgets for agent workflows ↗
A multi-step agent needs a budget for both resources and waiting time.
AI & automation · Article
Writing acceptance criteria for an AI-assisted feature ↗
Write reviewable criteria for evidence, uncertainty, failure behavior and action boundaries in an AI-assisted product feature.
AI & automation · Article
The unit economics of an AI feature in a commerce tool ↗
The unit cost of an AI feature is not simply the price of one model response.
AI & automation · Article
Evaluating AI merchandising recommendations offline ↗
Offline evaluation of AI merchandising recommendations asks whether a proposed system makes defensible suggestions on a fixed set of cases before customers are exposed to it.
AI & automation · Article
Designing observability for a multi-step commerce agent ↗
Observability for a multi-step commerce agent should let an operator reconstruct the work without guessing from the final answer.
AI & automation · Article
A threat model for agents reading merchant websites ↗
An agent that reads merchant websites crosses a trust boundary whenever page content influences its next action.
AI & automation · Article
MerchNivo: an evaluation plan for a store-audit workflow ↗
Define reference cases, separate detection from explanation, test merchant decisions and set release gates before making performance claims.
AI & automation · Article
Handling multilingual catalog ambiguity with AI ↗
Multilingual catalog work involves more than translating a product title.
AI & automation · Article
Evaluating an agent with replayable operational tasks ↗
A replayable operational task gives an agent a known starting state, a defined goal and a verifiable expected outcome.
AI & automation · Article
Building a ground-truth set for store-audit explanations ↗
A ground-truth set for store-audit explanations should connect each expected finding to observable evidence.
How I approach the work
Choose automation around a defined problem. Establish data quality, decision boundaries, human review and evaluation before expanding autonomy.