THE BOUNDEV JOURNAL
The engineering
archive.
Earlier writing on software, hiring, and building AI. Articles retain their original URLs and publication dates; some historical offers and references are outdated.
Browse the archive
LLM service tiers: which AI features belong on flex vs priority
Flex, standard and priority tiers sell the same model at different speeds and prices. Here is how to map each AI feature to a tier, with the fallback code that keeps flex
Agents need documentation more than memory: what to write down
Most production agents fail for lack of a glossary, a procedure, and a tool contract, not for lack of memory. What agent documentation looks like, and the narrow job memo
Your AI feature passed on demo data. Test it on production data
Demo data is clean and chosen by whoever wrote the prompt. Here is how to build a production-shaped test set in two days so the launch week is quiet.
Sync, async, or precompute: where an AI feature should run
Most AI features ship as a blocking model call inside the page request. Here is how to choose between synchronous, async job, and precompute-at-write patterns using three
Serverless GPU cold starts: the LLM cost nobody budgets
Serverless GPU inference looks cheap until cold starts inflate the bill. Here is the startup tax, the keep-warm tradeoff, and when dedicated wins.
Estimate an AI feature's LLM bill before you ship it
A back-of-envelope model to forecast an AI feature's monthly LLM cost before you build it, plus the four line items most estimates leave out.
Stop giving your production AI agent a god-mode API key
An AI agent that takes real actions needs its own scoped, short-lived identity, not a copy of your admin key. The pattern for agent credentials in production.
Timeouts, retries, and fallbacks that keep an AI feature up
A demo that works on your laptop still falls over in production. Here are the timeout, retry, circuit breaker, and degradation patterns that keep an AI feature up.
How to meter and bill AI feature usage without hurting margin
Metered AI features tie revenue to real token cost. Here is how to meter usage, pick a billable unit, and protect gross margin when you ship one.
Trajectory vs outcome: why agents pass evals and still fail
Outcome evals tell you an agent failed; trajectory evals tell you why. How to score the path an agent takes, and which eval gates a release.
GO FROM THEORY TO PRACTICE
See an expert review, step by step.
Explore a sample report, a review rubric, and the evidence behind an evaluation.