Field Notes

Notes from shipping real AI systems

No think-pieces. No hype. Just what we’ve learned from scoping, building, and maintaining AI agents for operators who can’t afford to waste a quarter on a stuck project.

Jul 12, 202612 min read

GPT-5.6 vs Claude Fable 5: Benchmarks vs Reality

GPT-5.6 Sol vs Claude Fable 5, with the numbers behind the noise: Fable leads SWE-bench Pro 80.3% to 64.6%, Sol wins Terminal-Bench at a third of the cost, and METR flagged Sol for the highest benchmark-cheating rate it has ever measured. A buyer's guide to reading the July 2026 scoreboard.

Jul 12, 202610 min read

GPT-5.6 vs Claude Opus 4.8: Do You Need the Frontier Tier?

GPT-5.6 Sol scores 59 on the intelligence index; Claude Opus 4.8 scores 56 and costs less per output token — and Opus wins SWE-bench Pro 69.2% to 64.6%. A buyer's guide to the most underrated question in AI procurement: when the workhorse tier beats the frontier tier.

Jul 12, 202611 min read

GLM-5 vs MiniMax M3: Open Models Got Serious

GLM-5.2 vs MiniMax M3 with real numbers: GLM leads Terminal-Bench 81% to 66% and tops the open-weight leaderboard; M3 costs 3.7x less on output, reads images and video, and beats Claude Opus 4.7 at autonomous web browsing. A buyer's guide to the open-weight generation that finally competes.

Jul 12, 202611 min read

GLM-5 vs Claude Fable 5 vs GPT-5.6: The Real Matchup

GLM-5.2 vs Claude Fable 5 vs GPT-5.6 Sol with real numbers: Fable leads SWE-bench Pro at 80.3%, Sol wins Terminal-Bench, and the open-weight GLM-5.2 lands within striking distance at 11x cheaper output. The open-vs-closed decision, benchmarked.

Jul 12, 202612 min read

Claude vs ChatGPT in 2026: Which One for Real Work?

Claude vs ChatGPT in 2026, from a team that builds on both: ChatGPT is the versatile all-in-one — voice, images, custom GPTs, an $8 entry tier. Claude is the work engine — better writing, deeper coding, stronger long-document analysis. Which to pick by use case, with the model-level numbers behind it.

Jul 12, 202611 min read

Claude Code vs Codex vs Cursor: The 2026 Field Test

Claude Code vs OpenAI Codex vs Cursor in 2026: terminal agent vs cloud async agent vs AI-native IDE. Pricing, context-window reality, the models underneath, and why most serious teams run more than one. From a team that ships client work with all three.

May 17, 202613 min read

AI Automation Agency vs AI Development Agency: What's the Difference?

AI automation agencies ("AAA") and AI development agencies are two distinct businesses serving different buyers. We break down what each one ships, who they hire, where the line really is, and which one matches your project — with a 5-row comparison table and a 4-step decision framework.

May 17, 202616 min read

What Is Retrieval-Augmented Generation? A Buyer's Guide to RAG in Production

Retrieval-augmented generation (RAG) is how most production AI systems answer questions over your data without hallucinating, retraining the model, or burning the context window. Plain-English explainer for non-technical buyers — what RAG is, when to use it, and the four failure modes that kill RAG projects.

May 16, 202611 min read

OpenClaw: The Personal AI Agent That Actually Does Things

OpenClaw is an open-source personal AI agent that runs locally on your own machine, integrates with 50+ tools (Gmail, GitHub, Slack, Telegram, browser, shell), and can write its own skills. We break down what it does, how it works, who it is for, and what it takes to deploy it well.

May 16, 202610 min read

Hermes Agent by Nous Research: The Agent That Grows With Your Server

Hermes Agent is an MIT-licensed autonomous agent from Nous Research that runs on your server, remembers what it learns, writes its own skills, and orchestrates isolated subagents. We unpack the architecture, the capabilities, and the practical work it takes to make a deployment pay off.

May 16, 202614 min read

Best Generative AI Consulting Companies (2026)

An honest, opinionated comparison of 10 generative AI consulting companies in 2026 — boutique practitioners, mid-sized AI specialists, Big 4 strategy firms, and enterprise giants. Includes who each one is actually best for, with a 10-row summary table.

May 16, 202612 min read

How to Build an AI Agent (Without an ML Team)

Plain-English guide to building an AI agent without an ML team. What an agent actually is, the five parts you need (model, tools, memory, control loop, observability), what you can realistically DIY, and the common mistakes that kill projects before week three.

May 12, 202613 min read

AI Agents vs. Traditional Automation: When to Use Each

AI agents and traditional automation solve different problems. Pick the wrong tool and you either overspend on an agent for a workflow Zapier could handle, or you cap the value of a workflow that needed judgment. A practical decision guide.

May 7, 20269 min read

What Drives the Cost of Building an AI Agent? A 2026 Honest Breakdown

Most cost guides for AI agents are generic brackets dressed up as advice. The real question is what drives the cost of your specific build. Four variables move the number more than anything else, and the LLM bill is almost never the expensive part.

May 7, 202611 min read

How to Choose an AI Development Agency: 12 Questions That Separate Real Builders from Hype Shops

If you are evaluating AI agencies, twelve questions are enough to separate the real builders from the hype shops. We are an agency. Here are the questions we wish more clients asked us — and the answers that should make you walk away.

May 7, 202610 min read

AI Agency vs In-House AI Team: A Decision Framework for 2026

Most AI-agency-vs-in-house posts are vendor advocacy in disguise. Here is a framework that gives the honest answer for your situation — including the specific cases where in-house wins, even when an agency would be cheaper this quarter.

May 7, 202610 min read

LangChain vs CrewAI vs AutoGen: What Business Buyers Need to Know Before Hiring an AI Agency

LangChain, LangGraph, CrewAI, AutoGen — the framework comparison posts are written for engineers, but business buyers are the ones writing the checks. Here is what each framework means for your timeline, your budget, and your ability to switch teams later.

Apr 7, 20267 min read

Why Most AI Projects Fail (And It's Not the Models)

Every few months a client tells us their AI thing is broken and asks if we can swap in a better model. We almost never touch the model. Here's what's actually going wrong.

Apr 7, 20269 min read

Build vs Buy for AI Automation: A Decision Framework

Every founder and ops lead we talk to has the same question: do we buy the AI tool or build our own? Here's the framework we use — and the uncomfortable fact that the right answer is often a hybrid neither side likes.

Apr 7, 20268 min read

How to Scope an AI Agent Project Without Getting Burned

The scope document is where most AI agent projects are lost — before anyone writes a single line of code. A field-tested checklist for operators who don't want to hand over a blank check.

Talk to us

Have a harder question?

The posts above cover the common patterns. The interesting problems are the ones that don't fit the pattern. Book a free strategy call and we'll dig into yours.

or schedule an AI consultation →