Skip to content

Data story Services Case studies Blog Process About Start a project

BLOG

Notes from shipping AI in production.

What we built, what broke, and what we would do again.

Horizontal bar chart of estimated cost per 1,000 chats for seven LLM models, from under one dollar to thirty-five dollars

AI Assistants & Search

LATEST · 20 Oct 2026

LLM API pricing in October 2026 with cost per 1,000 chats

List prices from Anthropic, OpenAI, Google and DeepSeek, turned into one comparable number, with the caveats and price changes still to come.

20 Oct 2026 · 5 min read

Diagram splitting a Next.js upgrade into two tracks, an immediate security patch and a later Cache Components migration

AI-Native Products20 Oct 2026

Upgrading to Next.js 16.4 without breaking production

Patch the October 14 security release first, then adopt Cache Components on your own schedule. A decision framework for Next.js 15.5 and 16.x apps.

Diagram showing four vector stores placed along an axis from low to high queries per month

AI-Native Products16 Oct 2026

Choosing a vector store for RAG by query volume, not vector count

S3 Vectors, OpenSearch, pgvector and Pinecone each win a different band of query volume. How to pick one for a RAG system and when to revisit it.

Diagram of a scraper pipeline with a fetch ladder, schema validation and a drift alert feeding a clean dataset

Data & Machine Learning13 Oct 2026

How to build a web scraper that survives site changes

A fetch ladder, schema checks and drift alerts that keep a production scraper from silently returning bad data after the target site changes.

Diagram of three teams and their AI assistants connecting through one MCP server to prompts, data and internal tools

AI Agents & Automation12 Oct 2026

Six decisions to make before you build an MCP server

Scope, transport, auth, versioning and logging choices that decide whether an internal MCP server gets adopted or quietly abandoned.