Blog
Insights on AI-native development and building SaaS products
AI Evals: How to Test AI Features Without a Research Team
Your unit tests mock the model, so they can't see the layer users actually touch. How to test AI features with golden datasets, a binary judge, and a CI gate that costs about $0.56 a run.
Why AI App Builders Hit a Wall — and What to Do Instead
Lovable, Bolt, and Replit are validation machines that stall at production. Where the wall sits for each, what migrations really cost, and how to keep AI speed on a foundation you own.
AI Agent Examples: 6 Real-World Use Cases to Build in 2026
Six real-world AI agent examples that work as businesses today: support, sales, data, ops, and research, with the exact tools, data model, and approval flow to build each.
Loop Engineering in Claude Code: Let the Agent Run Itself
Loop engineering means designing the system that runs your agent in a cycle. Drive a feature to done with /goal, run chores on a schedule with /loop and Routines.
When We Finally Added RAG — and Why pgvector, Not Pinecone
We argued most AI agents don't need RAG — then we shipped it. Where retrieval actually earns its place, and the copy-pasteable pgvector build that made it boring.
Human-in-the-Loop AI Tools in Next.js
Let your AI take real actions without going rogue. The human-in-the-loop approval pattern for AI write-tools in Next.js, with AI SDK v6 code and audit trails.
What Is Context Engineering? A Developer's Guide for 2026
Context engineering is filling the AI's context window with the right information before it writes a line — not better prompts. What it is, the four sources, and how to build it.
Frontier Models Got Smarter. Your Code Didn't Get Safer.
AI models got a generation smarter while AI-generated code's security stayed flat at ~55%. The 2026 data behind the industry's shift from velocity to verification.
AGENTS.md for SaaS: The README Your AI Agent Actually Reads
AGENTS.md is the file your AI coding agents read before they touch your code. What to put in one for a SaaS, what to leave out, and what the research actually shows.
Best AI Agent SaaS Tech Stack in 2026
An AI agent SaaS stack is a normal SaaS shell plus an agent layer. Our 2026 pick for each layer: framework, LLM gateway, orchestration, MCP, memory, AGENTS.md, and evals.
Do You Need RAG for Your AI Agent? Most Don't
Most SaaS AI agents don't need a vector database. The simpler architecture: file-based memory à la Claude Code's MEMORY.md, plus 1M-token context windows and tool calls.
Free vs Paid SaaS Boilerplates: What You Actually Get
Free SaaS boilerplates ship with 14k+ stars but stale repos. Paid kits cost $199–599 once. Here's the real cost-of-ownership math for both, with 2026 data.