The Depth Meter. Before you read a word, the amber bands show how much of a lesson is production reality and failure modes — the part nobody else writes.
AI Engineering Guide
Depth
The depth under the demo.
A guide that takes working software engineers to competent AI engineers — and keeps going past the point the demo works, into production reality, failure modes, cost, and latency.
A glimpse of what's coming
Every lesson does what a blog post won't.
Build it · stdlib
def allow(bucket, now):
bucket.tokens += (now - bucket.t) * RATE
bucket.tokens = min(CAP, bucket.tokens)
bucket.t = now
return bucket.spend(1)from scratch
Use it · a shared store
allowed = limiter.hit("tenant-42")the library
Differs in: whether the count is shared across replicas
Build it, then use it. The from-scratch version and the library doing the same thing, side by side — so the framework stops being magic. Toggle it above.
Leave with something you keep. Every lesson ends with a downloadable, MIT-licensed artifact — an eval harness, a runbook, a config — you can run at work on Monday.
Launching soon
Server-rendered, no signup, everything on the page.