Blog
News, guides, and updates from Curtis Foster.
How a non-technical founder can run an AI product day to day
You do not need to become an engineer to run an AI product well. You need clear operations, visible failure points, and a small system you can trust.
When to fine-tune, when to use RAG, and when to write a better prompt
A practical AI decision framework for choosing prompts, retrieval, or fine-tuning without wasting time, money, or engineering effort.
The startup metric almost nobody instruments: retry storms
A retry storm can turn one failed AI request into a four-figure bill. Learn how to spot the pattern, measure it, and protect AI reliability.
Zero to one with generative image and video pipelines using Imagen, Veo, and FFmpeg
A practical guide to building a reliable AI video pipeline with Imagen, Veo, and FFmpeg, from the first still image to a finished export.
Prompt engineering is table stakes. System-prompt engineering is the real skill
Prompt engineering can make a demo work. System-prompt engineering makes an LLM product behave reliably when users, tools, and edge cases arrive.
A founder’s guide to avoiding AI vendor lock-in without slowing down
You do not need to build every AI component yourself. You need a small model abstraction, clean boundaries, and a plan for switching before switching becomes urgent.
How I ship a new AI product in a weekend and the three things I refuse to skip
A practical weekend process for building an AI MVP without confusing speed with recklessness. Ship fast, learn in public, and keep the safeguards that protect users.
AI guardrails aren't a filter. They're an architecture
A production AI product needs controls around data, tools, models, users, and recovery paths. One moderation filter cannot carry that load.
Multi-model orchestration: why I stopped betting a whole product on one model
A production AI product should not depend on one model's price, latency, or behavior. Here is how I use routing, fallbacks, and evaluation to spread that risk.
The real cost of an AI feature: instrument AI cost before it surprises you
An AI feature costs more than its model call. Learn how to measure retries, tokens, latency, storage, and AI unit economics before your bill gets ahead of you.
What long-term memory actually takes in a production LLM application
LLM memory is not a bigger context window. It is a data system with rules for capture, retrieval, correction, privacy, and cost.
RAG is a data-quality problem, not a search problem
Most RAG failures begin before vector search runs. Fix the source data, ownership, and update process, and retrieval becomes far easier to trust.
Subscribe to Curtis Foster
Get new posts in your inbox. No spam, unsubscribe anytime.