Writing on production AI.
We write about the work of putting AI into production and keeping it there: agents, retrieval, document automation, integration and MCP, evals, and cost control, with AI security in preparation. Every piece is anchored in real implementation detail, and every load-bearing claim carries its source.
The four questions to ask before you buy AI automation
You do not need to understand AI to buy it well; you need four answers a vendor should give in plain words.
document automation
RAG that cites its source and says when it does not know
A knowledge assistant earns trust with two behaviors no model provides by default: every answer carries a citation that resolves to the exact passage it rests on, and a question the corpus cannot answer gets a plain refusal instead of a fluent guess.
rag
What an MCP server does to your guardrails
A guardrail is in force only at a chokepoint that every execution path passes through, and adding an MCP server adds a path.
mcp
Which documents should skip human review?
A document skips human review when two conditions hold: every consequential field on it can be checked against a second source the business already holds, and every action it triggers can be reversed.
document automation
The model was never the bottleneck
What stops a stalled pilot is a short list of missing mechanisms, and the model is not on it.
llm integration
Why your AI pilot did not reach production
Your AI pilot did not reach production because it was built to prove a model works, not to survive an operation.
document automation
Is 95% accuracy good enough, and who signs it off?
No percentage is good enough on its own, and the sign-off belongs to the owner who carries the consequence of a wrong value, not the engineer and not the vendor.
evals
Does the human review gate eat the savings?
A review gate pays for itself when the time saved on the items that pass exceeds the time spent on the items that stop.
cost control
What we check before an AI agent touches a real system
Before an agent we build gets access to a real system, it passes a fixed preflight: bounded scope, a human approval gate, an eval set it must clear, a kill switch, an audit trail, a cost ceiling, and a tested rollback path.
agents