Field noteWhy your RAG demo lies in production
The retrieval that looked flawless on ten test questions, and what actually happened when real users asked the messy ones — plus the three fixes that mattered most.
Issue 12
← Resources
Newsletter · monthly dispatch
One sharp read a month on shipping AI for real.
A monthly dispatch on shipping AI in production — what's actually working, what's hype, and the engineering decisions behind both. Written by the team that builds it, for the people who have to ship it. No spam, no fluff, unsubscribe in a click.
monthly/practical/no spam/unsubscribe anytime
Scroll
01Why subscribe
AI moves fast. Most of the noise isn't worth your time.
There's no shortage of AI takes — threads, launches, breathless predictions. Very little of it survives contact with a real deadline and a real production system. This dispatch is the opposite of that firehose: once a month, a few honest field notes from actually shipping AI, an unsentimental read on what's genuinely useful versus what's a demo, and a practical thing or two you can use. It's written by the engineers doing the work, it earns its place by being worth reading, and it's one click to leave whenever it stops being so.
02This month
Issue 13~10 min read
What actually broke when we shipped AI to production.
A typical issue, in other words. This month: the latency budget nobody planned for, the prompt change that quietly doubled costs, and the evaluation harness that caught a regression before users did. Plus an honest take on a trend everyone's overhyping, and a one-page checklist from the kit. Practical, specific, and free of the “AI will change everything” throat-clearing.
Read a sample issue03In every issue
Ten minutes, worth reading.
- 01
A field note
One real lesson from shipping AI this month — the kind of thing you only learn after it breaks.
- 02
Signal vs hype
An honest take on a trend everyone's talking about: what's genuinely useful, what's just a good demo.
- 03
Something we shipped
A concrete build, with the decision behind it — the trade-off, not just the result.
- 04
A tool or checklist
A practical file you can use today — a scorecard, a checklist, a template, straight from real work.
- 05
A few links worth it
A short, curated handful — the pieces actually worth your time, not a firehose of tabs.
- 06
Reply welcome
Hit reply with a question or a war story. We read every one, and the best ones shape future issues.
Signal, not noise.
One considered issue a month — the opposite of the AI firehose.
04Recent issues
A taste of what lands.
Field note
Signal vs hypeAI agents: what's real, what's a demo
An honest read on where autonomous agents are genuinely useful today, where they quietly fall apart, and the boring guardrails that decide which side you land on.
Issue 11
From the kitThe AI readiness scorecard
A one-page tool from a recent issue: twelve honest questions for deciding whether an idea is actually ready to build with AI — or whether plain software would win.
Issue 10
05The deal
A fair trade for your inbox.
Monthly, not daily
One considered issue a month — never a stream of nudges, and we'll skip a month before we pad one.
No spam, ever
Your email stays with us — no third parties, no selling, no sneaking you onto other lists.
One-click out
Unsubscribe in every issue, no guilt-trip, no 'are you sure?'. We'd rather you stay because it's good.
One good email a month.
Join the dispatch on shipping AI in production — signal, not noise.
Subscribe06Questions
The things people ask first.
A monthly dispatch on shipping AI in production — one focused read, not a daily firehose. Each issue is a few field notes from real work: what we shipped, what actually broke, what's genuinely useful versus what's just hype this month. It's written by the engineers doing the work, for people who have to ship it themselves.
Ten minutes, once a month.
Subscribe to the dispatch on shipping AI in production — honest field notes, a signal-vs-hype read, and a practical tool or two, from the engineers doing the work. No spam, and one click to leave whenever you like.