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 Articles retain their original URLs and publication dates; some historical offers and references are outdated.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
Why pure vector search misses the queries that matter
Pure vector search drops exact terms, IDs, and numbers. Here is how to combine BM25 and dense embeddings with RRF and reranking for better RAG recall.
When to collapse a multi-step agent into one reasoning call
Reasoning models can collapse over-built agent chains into a single call, but they overthink. Here is how to decide which tasks to simplify and which still need an explic
What hiring an AI engineer really costs your cap table
Founders model the salary and forget the equity. A senior AI hire can cost 0.5-1.5% of the company. Here is how dilution compares to a subscription you can cancel.
Your AI engineer starts Monday. When do they actually ship?
Time-to-hire ends on day one. The ramp from start date to a shipped AI feature usually runs a full quarter. Here is what it costs and how to shorten it.
Why your AI agent's memory rots in production (and the fix)
Agent memory rots when you only write and never forget. Design the write, retrieve, and forget paths so production answers stay accurate.
Why your LLM bill is unpredictable (and how to fix it)
A provider invoice tells you the total, not the cause. Build a per-feature tagging layer so you can see which features, customers, and workflows drive your LLM spend.
Cut your LLM bill in half with the Batch API
Both OpenAI and Anthropic price batch endpoints at half the standard rate. Here is which of your LLM workloads qualify and how to ship batching without a rewrite.
Why your AI agent fails on tools, not the model (and the fix)
Production AI agents fail at the tool layer, not the model. Design function-calling tools like API products: narrow contracts, validation, idempotency keys, structured er
Why your LLM keeps returning broken JSON (and the fix)
Plain JSON-mode prompting fails 8 to 15 percent of the time. The fix: pick the right enforcement level, validate for meaning not just shape, and ground extraction in retr
Keep your AI feature up when the LLM API fails
Every LLM provider has an outage eventually. Retries with jitter, cross-provider fallback chains, and circuit breakers turn a provider failure from an incident into a non
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.