Small Pond Software, LLC

Notes from the pond

Articles

Notes on building focused, practical software, written as we build it.

Large number 5 inside concentric ripples next to the title, Five questions to ask before you hire an AI builder.

October 14, 2026

Five Questions to Ask Before You Hire Anyone to Build AI for Your Business

Five questions to ask any vendor before you pay for AI automation: logs, account ownership, tools, where your data goes, and a process map.

Read article

A search results list where three imposter results rank above the right patent, which has slipped to fourth, with the note: no error, no warning.

October 5, 2026

After 1.2 Million Patents, Nobody Told Me My Search Was Drifting

A search that put the right patent first started burying it, with no error and no way to prove it. Why I built an eval suite into my new private chat app.

Read article

Headline The hidden discount on your business, with a price tag marked 15 to 25 percent off

October 1, 2026

The Hidden Discount on Your Business: Why Buyers Pay Less When You're the One Holding It Together

Buyers often pay 15 to 25 percent less for a business that depends on its owner. Here are the red flags they look for, and how to fix them before you sell.

Read article

Autonomy, competence and relatedness, the three needs of Self-Determination Theory, with the question: will AI be a replacement or an amplifier?

September 22, 2026

AI and Self-Determination Theory: Will AI Make Us More Capable or Less?

How AI can strengthen or weaken autonomy, competence and relatedness, what the Industrial Revolution teaches us about disruptive technology, and why AI works best as an amplifier, not a replacement.

Read article

Five-step automation framework: Log It on Monday, Score It on Thursday, Write It Up on Thursday, Test It on Friday, Decide on Friday.

September 18, 2026

What to Automate First: A Small Business Automation Framework

A five-day framework from a Colorado Startup Week talk for scoring your first business automation before you commit, plus the five-stage maturity map for getting there.

Read article

Diagram of the Navier-Stokes incident as a pipeline: mathematicians' private work flows into OpenAI's Codex tool, through an unknown pathway, and out as a breakthrough proof, with OpenAI's statement that it cannot rule out that usage data helped improve its models.

September 10, 2026

Why Local AI Matters More Than Ever

Why sending your data into someone else’s AI model is a competitive risk, and what running models locally with Gemma, Ollama and SentenceTransformers is like.

Read article

Two panel diagram comparing where project status comes from: scattered, self-reported sources like requirements, dev status, risks, testing, and schedule, versus evidence signals like commits, tests, transcripts, requirement changes, and deployments feeding an AI-drafted picture that a person still decides on.

September 7, 2026

Who Owns This Task When AI Agents Do the Work?

We’re adding AI agents to project teams while still tracking status with tools built for human self-reporting. A large ERP program shows why evidence beats self-reported status, and where AI fits without replacing human judgment.

Read article

Comparison graphic of three AI system design paradigms: Single LLM, Structured Workflow, and Autonomous Agent.

August 27, 2026

Comparing AI System Designs

A staffing analogy for choosing between single LLM calls, structured workflows, and autonomous agents, and why the cheapest tool that solves the problem beats the most impressive one.

Read article

Title card contrasting semantic RAG search, the retrieval layer, with AI agents, the orchestration layer.

August 14, 2026

AI Agents and Semantic RAG Aren't the Same Thing

A customer service refund request shows why RAG retrieves trusted context while agents coordinate the decision and the action, and why mixing the two up gets teams in trouble.

Read article

Building Council of AI: four AI models, one question, and the bugs hit getting them to agree.

August 14, 2026

Building Council of AI

Four AI models, one question, and two production bugs that shaped how the pieces fit together.

Read article

Hand-Rolled Chunker vs. LangChain Splitters comparison for chunking patent text in a patent ingestion pipeline.

August 14, 2026

Why I Didn't Use LangChain to Chunk Patent Text

We skipped LangChain for the chunking layer in our patent ingestion pipeline and wrote our own splitter instead. Here's why that was the right call.

Read article