Field notes on production AI.
Long-form essays on what actually ships in production AI engineering — advanced RAG, agents, fine-tuning, evals, and the boring infrastructure that makes them all work. Written in the same voice we use in proposals.
- Production AIApril 12, 2026·14 min readPillar essay
Production AI: What It Actually Takes to Ship LLM Software That Doesn’t Wake You Up at 3 a.m.
A working demo is not a working system. Here is the practitioner playbook for getting LLM applications into production without the usual midnight failure modes.
Read article - Engineering RescueMarch 25, 2026·11 min readPillar essay
When to Rebuild, When to Rescue: A Senior Engineer’s Triage Framework for Failing Codebases
A practical framework for deciding whether to save the codebase you have or start over — and how to make either choice without losing the next two quarters.
Read article - Fractional CTOFebruary 18, 2026·9 min readPillar essay
Fractional CTO Services: When You Need One, How to Hire One, What to Expect
A founder’s guide to engaging a senior engineering leader part-time — the situations where it works, the situations where it does not, and what to look for in the contract.
Read article - Production AIApril 26, 2026·8 min read
Why your LLM app needs an eval harness before it needs a frontend
The single highest-leverage piece of infrastructure in production AI is the eval harness — and most teams build it last, after they have already shipped problems they cannot detect.
Read article - Production AIApril 19, 2026·10 min read
RAG in production: eight failure modes nobody tells you about
The naive RAG pipeline that demos beautifully fails in eight specific, predictable ways once real users hit it. Here is the field guide.
Read article - Production AIApril 12, 2026·9 min read
Building AI agents that don't drift: a practitioner's playbook
Multi-step agents are the highest-value and most failure-prone class of AI software. Three principles separate the ones that ship from the ones that get rolled back.
Read article - Production AIApril 5, 2026·8 min read
How to keep your AI bill under control without crippling the product
Six concrete patterns we apply on every cost-optimization engagement — typically cutting bills 40–80% with no behavior change.
Read article - Production AIMarch 29, 2026·7 min read
Claude, GPT, or Gemini for production? A working engineer's decision tree
Skip the vendor benchmark wars. Here is how we actually pick a model for a production engagement, with the criteria that matter and the ones that do not.
Read article - Production AIMarch 22, 2026·7 min read
Human-in-the-loop patterns for AI products that can't afford mistakes
Three design patterns for inserting human review into AI workflows without killing the throughput that made you want AI in the first place.
Read article - Production AIMarch 15, 2026·8 min read
Observability for AI: what to log, what to alert on, what to ignore
Conventional observability watches for things going wrong. AI observability has to also watch for things slowly going less right. Here is the practical setup.
Read article - Engineering RescueApril 22, 2026·9 min read
The anatomy of a vibe-coded codebase: 12 signs you're holding technical debt disguised as speed
Codebases generated largely by AI assistants without a senior reviewer in the loop have a recognizable shape. We rescue them every quarter.
Read article - Engineering RescueApril 8, 2026·7 min read
Why your AI MVP won't make it to production (and what to do about it)
The pattern we see most often: a working AI demo that has been "almost ready to ship" for nine months. Five reasons why, and how to break the loop.
Read article - Engineering RescueMarch 25, 2026·6 min read
The first two weeks: a checklist for inheriting someone else's codebase
A 14-item checklist we run on every Engineering Rescue diagnostic. Free to copy and paste into your internal docs.
Read article - Engineering RescueMarch 18, 2026·6 min read
Five signs it's time to replace your development agency
Knowing when a vendor relationship has gone south is harder than it should be. Five concrete signals — when you see two or more, it's time.
Read article - Engineering RescueMarch 11, 2026·6 min read
Migrating a JavaScript codebase to TypeScript without losing the quarter
Five rules for a TypeScript migration that ships incrementally instead of becoming a six-month rewrite that misses two product cycles.
Read article - Engineering RescueMarch 4, 2026·7 min read
Why one senior engineer beats three juniors for high-stakes software
The math seems wrong: 3× the headcount should produce more output. For high-stakes software it does not. Here is why.
Read article - Fractional CTOApril 15, 2026·7 min read
Fractional CTO vs. dev agency vs. full-time hire: a cost-and-risk comparison
Three ways to add senior engineering capacity, three different risk profiles. Here is how to pick the one that fits your situation.
Read article - Fractional CTOApril 1, 2026·8 min read
AI strategy for non-technical founders: a 90-minute mental model
Six concepts that let a non-technical founder make sound AI decisions without becoming an engineer. Read this before any vendor pitch.
Read article - Fractional CTOMarch 28, 2026·6 min read
The embedded engineer model: why top talent is moving away from project work
Senior engineers increasingly prefer multi-month embedded engagements over short projects. Why this matters for how you hire.
Read article - Fractional CTOMarch 21, 2026·7 min read
Build vs. buy for AI: a decision framework you can actually use
Five questions that resolve the build-vs-buy decision for AI features without the marketing fog. Use it before any vendor demo.
Read article - Fractional CTOMarch 7, 2026·7 min read
What engineering looks like at Series A: a survival guide for founders
A working framework for what your engineering org should look like in months 0–18 after Series A. Hire wrong here and you spend year two cleaning up.
Read article
Like the writing? You’ll get the same on your engagement.
Every Mini Trends engagement comes with written architecture memos, weekly progress notes, and a final handover document — written in the same voice you just read.