Articles

AI Fact Checking: How I Check My Content Reliably With Agents

At the end of September I had AI agents fact-check all 64 articles on this site, one agent to an article, and their reports came to 155 fixes. Google's own guidance, updated on 1 October, now calls it critical to fact-check AI-generated content before you publish it. In today's post we're taking a closer look at how the agents were set up, what they found, what the recent research says about AI fact-checking, and how I got the fixes onto live pages without breaking them.

claude-code · AI agents

OpenCode vs Claude Code: Setting Up OpenCode Desktop on Windows

OpenCode Desktop runs whichever model you point it at, and it picked up my Claude Code skills without being asked. I connected DeepSeek, GLM-5.3 and a local Qwen, fenced the permissions, and then had GLM-5.3 review an MCP server I'd built with Claude, fix it and get a release ready.

claude-code · mcp

How to Connect Google Trends to Claude Code

Google Trends can now be read by whole category, with no keyword, so I added it to my seo-audit MCP. In today's post we're connecting Google Trends to Claude Code, running the three ways of asking on real data, and having Gemini MCP draw the report, with every prompt included.

claude-code · mcp

Claude Code Skills: What They Do and Which Ones to Use

A skill isn't an extension to the system prompt, and it doesn't do the job of an MCP server. It shapes what Claude produces without you spelling out what you want every time, and in this post I go through where a skill sits in a call, the skills worth using (my two included) and a template to build your own.

Claude · claude-code

CLAUDE.md: How to Write One That Stays Lean

In today's post, I'm taking a real CLAUDE.md from one of my projects down from 180 lines to 67. The review I ran on it came back with 23 findings, four of them serious, and I'm sharing the prompt so you can run it on your own repo.

Claude · claude-code

How to Make a PDF Report with Claude

In today's post I'm showing how I make a monthly PDF report in Claude from a CSV, and why I think it's a perfectly legitimate way to report, provided Claude checks every number against the file. A house-style Skill sets the look, and the outline is agreed before anything is built.

Claude · Automation

How to Make a Presentation with Claude

In today's post I'm sharing how I make presentations with Claude, a way of working that keeps my time on the data and my insight rather than on PowerPoint. Claude writes the storyline first and puts every number in one data file, Gemini draws the charts, and the same source builds both a PDF and an editable PowerPoint file. When next month's numbers arrive, the deck rebuilds in under three minutes.

claude-code · gemini

How to Benchmark vLLM: Find the Best Model, Quant and Settings for Your GPU

In today's post I'm sharing how I benchmark local models on vLLM by changing one thing at a time, along with the traps that quietly skew the numbers and my results from 21 setups on two 48GB RTX 4090Ds. The harness, vllm-bench, is free, so you can run the same tests on your own card.

local-llm · vllm

Claude Code Memory: How to Keep It Useful

In today's post I'm going through Claude Code memory: how CLAUDE.md and auto memory work, why they go stale, and the lint and retire-don't-delete prune that keep mine useful, plus my designs for a scheduled skill and a small knowledge graph keyed to Wikidata.

Claude · claude-code

What is a Marketing Engineer (and do you need one)?

In today's post we're taking a closer look at the marketing engineer, the role Greg Isenberg reckons becomes tech's next big hire. The title may or may not stick, but the job is certainly real: I've been running the systems it describes for a year. Here's what it involves, what to build first, and what the sceptics get right.

Automation · AI agents

Infographics and articles that update themselves, built on free APIs

Once an article goes live, it starts to go stale, so I built The Daily Visual, a prototype site that rebuilds its charts, its data and the numbers in its sentences every morning from free public APIs. In today's post we're taking a closer look at how far that goes, what broke in its first month, and where I'd take it next.

claude-code · Automation

Complex ERP trade pricing on Shopify Plus B2B, native-first

We've put a global hardware distributor's ERP trade pricing live on Shopify Plus: about 70 hand-set price lists, with per-customer specials and volume breaks on top. In today's post we're taking a closer look at what native Shopify Plus carried, and the one gap in it we filled with a small first-party layer.

Automation · shopify-plus

vLLM settings for a pair of RTX 4090s: the flags I run for twelve local models, and why

The flag-by-flag vLLM settings page for a dual-4090 rig: the baseline every preset shares, the six flags that change per model, the rig discipline you set once, and one verified launch block for each of twelve open-weight models with the reason attached.

local-llm · vllm

The 26%: what the companies actually making money from AI do differently

Google Cloud surveyed 2,403 executives and found a 26% cohort whose AI returns are accelerating year on year - and BCG independently landed on the same number. Meanwhile MIT says 95% of pilots die. Both are true, and the difference between the two groups is where they aim the technology, not how much they spend on it.

Which AI pitches the best content ideas? Free Workers AI models vs DeepSeek

I cut my own content SaaS down to a free one-page tool to test demand, then ran the same brief through all eight AI engines it offers. In today's post we're taking a closer look at what each one pitched and how long it took, because on the free engines you pay in waiting.

Local LLMs · content marketing

KV Cache Explained: Why a Local LLM Won't Load on Your GPU

In today's post I'm taking a closer look at the KV cache, the memory a model keeps of the conversation so far, and why 'does it fit in VRAM?' is the wrong question when a local LLM won't load. It's what refused a 70B model its own 131k context on my 48GB card, and I'll show you how to size it, the vLLM settings that control it, and the kilobyte of GPU shared memory that stopped a 16GB model outright.

local-llm · vllm

RTX 4090 48GB: A vLLM Benchmark on Two Modded Cards

I built a local AI server at home from two modded 48GB RTX 4090 cards, £6,200 for the pair, and then spent about five days benchmarking them on vLLM, changing one thing at a time. These are the results, and the received wisdom about these cards that the numbers overturned.

LLM Inference Optimization: What Made Two RTX 4090s Faster

In today's post I'm going through what made local LLM inference faster on my two modded 48GB RTX 4090s, in the order it mattered: the file format, how you split a model across cards, locking the clocks at boot, speculative decoding and load times, with the numbers from my own testing.

local-llm · vllm

How to set up an MCP gateway with Docker

A lot of MCP servers get pulled straight off npm and run with your permissions, which makes them a supply-chain risk. In today's post we're taking a closer look at why I run mine in Docker containers, how to set up Docker's MCP gateway for Claude Code and Claude Desktop, and what a container won't stop.

Claude · mcp

How to Use the Gemini API (and Why I Run It Next to Claude)

Get a Gemini API key, make your first call in curl and Python, dodge the thinking-token trap that returns an empty answer, and see why running Gemini next to Claude is the real unlock. Written from production - and the bills.

Claude · mcp

Which AI is right for your job? ChatGPT, Claude, Gemini and Copilot, task by task

Everyone's talking about AI assistants - ChatGPT, Claude, Gemini, Copilot - and if you've a real job to get through, you mostly want to know one thing: which of them will build Monday's deck, sort the badly-exported spreadsheet, write up the meeting, and clear the inbox? Here's the answer, task by task.

Claude · gemini

Muse Glimmer 30B vs qwen: my day-one local benchmark on a dual RTX 4090 rig

Meta dropped Muse Glimmer 30B and I spent the day trying to unseat my qwen daily driver on the dual-4090 rig. Four apparent hangs, one day-one bug, and a same-harness bench later, I had my answer - and it wasn't the coronation I'd half expected. So, is it better than my daily driver?

gpu · hardware

How to Write a PRD an AI Can Build From (with a template)

A PRD is the difference between an AI coding tool that guesses and one that builds the thing you meant. Here's what a PRD is, a copyable six-part template, a worked example, and the two lines that do most of the work - written for the era where the thing reading your spec is an agent, not just your engineering team.

claude-code · PRD