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 Articles retain their original URLs and publication dates; some historical offers and references are outdated.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
SLOs for AI features: how to set targets a model can hit
How to define latency and quality SLOs plus an error budget for an AI feature, so you know when to ship a model change and when to hold.
What a senior AI engineer really costs: 9 months, not a salary
The fully loaded cost of a senior AI engineer hire in 2026, and the six to nine months to full productivity that the salary line never shows.
Provisioned throughput vs on-demand: when it saves money
Provisioned throughput reads like a discount, but it only beats on-demand above roughly 60-85 percent sustained utilization. Here is how to find your break-even in tokens
Stop AI agents burning your budget in a runaway loop
AI agents can burn 50x the tokens of a chat call when a tool-call loop fails to make progress. Here is where the cost comes from and the hard caps, loop detection, and pe
Context rot: why your AI feature gets worse as the prompt grows
Context rot drops LLM accuracy as your prompt grows, long before the token limit. How to detect it and design a production AI feature around it.
When your AI feature has a bad day: an incident runbook
When an AI feature degrades in production, a standard SRE runbook leaves you blind. Here is how to detect it, localize the failing layer, and mitigate fast.
The states your AI feature forgets: loading, error, empty
The wait, the failure, and the blank first screen decide whether users trust your AI feature. A practical guide to designing the states most teams skip.
Your production AI agent has too many tools (and the fix)
A growing tool list inflates token cost and drops tool-selection accuracy. How progressive disclosure keeps production AI agents fast and accurate.
Your RAG fails at the PDF, not the retriever
Most RAG systems fail at document parsing, not retrieval. How to parse PDFs, tables, and layout-heavy documents for RAG, and how to test extraction quality before you shi
Contextual retrieval: cut RAG failures without more infra
Anthropic's contextual retrieval prepends chunk-specific context before indexing, cutting top-20 RAG retrieval failures by up to 67 percent. What it costs and when it bea
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.