Autocomplete. Real World AI

I Would Not Mind Being Stuck on Opus 4.8 Forever

My real token bill for the last period should have been about $600,000. I paid $2,000.

The Cheapest Tokens in History, and the Biggest AI Bills

Per-token prices fell about 98 percent in three years. Enterprise AI spend tripled anyway.

HTML Is the Only Right Interface for Working Together With Agents

The Best Model Is the One You Know How to Use

Hot take:,Benchmarks do not matter to 99% of us. LinkedIn driven AI Brofluence is hurting your bottom line.

He Gave the Agent a Credit Card and Went to Bed

Twenty-four hours later its operator was holding a $6,531 cloud bill

Fable 5 Is the First Model You Have to Tell When to Stop

Anthropic just put a Mythos-class model in public hands. and the catch shows up on your invoice.

The AI Act Follows Your Output, Not Your Address

This is not just ‘European problems’ Doing business with Europe, deal with the EU AI Act.

The Small Agent That Actually Stuck

The boring one will win, every time.

The Smallest Eval That Beats an Opinion

You cannot verify a result against a standard you never wrote down.

AskWell Sends an Interviewer to Every Stakeholder, and Maps Where They Disagree

I built an app that replaces the form you send with an assistant that holds a real conversation.

Prompt Engineering, Summer 2026:

A field guide to what works, what stopped working, and what is genuinely new, with the numbers behind each call.

WTF Is a Agentic Coding Loop? And How to Build One.

They say stop prompting and start looping; Here is the practical version, with the commands.

The Right Argument About AI and the Environment

A consultant’s case for the honest middle, with the numbers checked.

Cost per Solved Task, Not Cost per Token

The number printed on the model’s price page used to be the number that mattered.

Claude Code Today

A working map of what Claude Code has become.

You Cannot Escape Your Way Out of Prompt Injections

A researcher who breaks AI systems for a living leaked a hospital president’s medications with one sentence. The reason this works is the…

The Loop Manual: From a While-Loop to a Self-Improving Fleet

A field manual for building agentic loops with Claude, from a bash while-loop you can paste right now to a self-improving fleet you run on…

What Anthropic’s AI Fluency Index Says About How We Actually Use AI

We get better at directing AI as we use it more but we get worse at doubting it, when the work looks finished.

With Agents, the Evals Are the Engineering

And Without Evals, the Engineering is Incomplete

Apple Killed the App Store. By Moving the Front Door.

Dramatic title aside, the real story here is the introduction of the second-surface.

The AI Act Didn’t Get Delayed the Way You Think

So relax, right?

Apple’s On-Device Apple Intelligence Replaced a Cloud API in My Desktop App

There is a capable language model already sitting inside macOS, free, offline, and very private.

Five Signals From the Frontier of AI Research

The things Frontier Labs are working on right now

Build a real forensic shoeprint matcher with me.

And Find out That the Bottleneck Is Not the Model

Your Agents Are in Production. That’s When They Get Rolled Back.

The second AI reckoning is not about cost. It is about control.