Getting things into production
What breaks between demo and live, what a shadow month actually buys, and why autonomy should be granted one task at a time.
Writing
Field notes, method write-ups and a few unpopular opinions. No gated PDFs, no email wall, and nothing written to make this work sound easy.
All writing
Field note
The demo runs on twenty hand-picked examples. Production runs on the twenty thousand nobody looked at.
Method
A single percentage hides the only distinction that matters in operations: mistakes you can absorb, and mistakes that reach a customer.
Opinion
Six situations where a scheduled job, a lookup table or a two-hour conversation beats anything we could build.
Practice
The agent does the work. Nobody uses the output. It is the highest-value month in any deployment and the first thing clients try to skip.
Economics
Token pricing is the smallest line in the bill. Here is the whole bill, with the parts vendors leave out.
Engineering
Six years of three different people naming things three different ways. What actually works, and what only works in benchmarks.
No gated PDFs, no email wall, and nothing written to make this sound easy — including the deployments that went badly.
What breaks between demo and live, what a shadow month actually buys, and why autonomy should be granted one task at a time.
Why one accuracy figure means nothing, how to split errors by type, and how to keep an evaluation set alive instead of letting it rot.
What an agent really costs to run, where the leverage sits, and when the honest answer is that this should not use AI at all.
Still have a question
All of this came out of real deployments. If your situation looks like one of these pieces, a thirty-minute call is usually more useful than another article.