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 Articles retain their original URLs and publication dates; some historical offers and references are outdated.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
Independent evaluation for AI features: who should grade your AI?
Why self-grading fails for AI features, what independent evaluation requires, and a practical setup SaaS teams can run every week.
What does an LLM evaluation cost? A budget model for SaaS teams
The four cost lines in an LLM eval run, a worked example you can copy, and where teams overspend on judges, human review and re-runs.
Review the plan, not the diff: a checkpoint for coding agents
The cheapest place to catch a bad coding-agent change is the plan. A plan template, a review checklist, who should review, and how to measure if it pays off.
AI coding agent spend per engineer: how to budget it
AI coding agent bills now vary 10x between engineers on one team. Measure per engineer, tie spend to merged work, and control cost through tiers and defaults instead of h
How to A/B test an AI feature in B2B SaaS without fooling yourself
AI features break standard A/B testing assumptions. Randomize by account, measure real work instead of clicks, and treat cost and latency as guardrails.
Deleting customer data from an AI feature: the full map
Once you ship an AI feature, a customer's data lives in prompt logs, vector indexes, caches, eval sets and provider-side objects. Here is the map and the deletion job des
Stop your AI feature from signing other people's names
Generative models attach real names, signatures, quotes, and logos to content those people never made. Four cheap checks that stop false attribution in SaaS AI features.
Giving an outside AI team data access without leaking customer data
An outside AI engineering team rarely needs raw production data. Here is the access setup by phase: owned keys, masked staging data, time-boxed read-only views, and a day
How to red-team your AI feature in CI before attackers do
Red-teaming an AI feature should not be a one-off pentest. Build a product-specific attack suite, grade it deterministically, and run it in CI on every prompt, model or t
Idempotent LLM jobs: stop paying twice for the same AI call
Queue redelivery, SDK retries, webhooks and double clicks all turn into paid duplicate LLM calls. Here is the idempotency key design and storage pattern that stops them.
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.