What is an AI agent harness? The loop behind Claude Code and Codex
What an AI agent harness does, with a short TypeScript loop. Where Claude Code, the Agent SDK, Codex and MCP fit.
Blog
Articles and guides on search, conversion, Shopify, software and AI agents. Each piece shows its date and author. They explain how the work is done; they aren’t case studies.
Articles
CVE statistics for 2026: each doubling came faster
I counted the CVEs published each month in the official list. Each doubled level came faster than the last: 12 years, 7 years, 2 years, then 5 months. Why the count is rising, what it doesn't mean, and how to decide what to patch first.
AI-generated unit tests: how do you know they check anything?
When an agent writes the code first and the tests second, the tests can describe its code, bugs included. Two patterns to look for, a small example you can run, and the red-green loop I use instead.
Zaraz consent: two questions instead of a cookie form
How Cloudflare Zaraz gates GA4, PostHog, Google Ads and Microsoft Advertising by consent purpose, why I replaced its dialog with a prompt that asks one question at a time, and how Preview & Publish and the cf CLI keep each change reviewable.
Affinity MCP vs code: why I make my thumbnails with a script
I wanted to show an agent making a thumbnail in Affinity through MCP, and it didn't work as expected. The same prompt as code worked much better. The prompt, part of the script, and why my timings aren't a benchmark.
AI coding security: the tunnel my agent opened without asking
Claude Code put my files on a public URL to get them into Google Slides, and didn't ask first. What that exposed, what to check before an agent can do the same, and why the responsibility stays with whoever connected it.
Denial of wallet: what a hard budget cap actually stops
Simon Willison argued for default hard budget caps, and I agree. What the current provider limits actually stop, what keeps billing anyway, and what to check before you deploy.
JavaScript number precision for large IDs, and ten new features for agent code
JSON.parse can turn a 19-digit ID into a different number without an error. How to keep IDs exact, then ten new JavaScript features for agent code, starting with that fix: what ships now, and what is still a proposal.
SSE vs WebSockets for AI apps: ten web standards and two patterns
Streaming a model’s answer one way? SSE is often enough. Here are ten web standards and two patterns for AI apps: what each is good for, where it falls short, and the spec to check.
Tailmanac: why I built a dog-care handoff app
When someone else looks after Darcy, my Papillon, they need her food, medicine, vet and routines, not her tricks. Tailmanac is the iPhone app I built to hand that over. Here's what it shares, why it shares an image, and what's still only an idea.
WebMCP vs MCP: when an agent should go through the browser
WebMCP gives an agent inside the browser structured tools on the page. A direct MCP server lets any agent call the service with no browser at all. I've built both. Here's when each one fits, with Shopify's checkout as the worked example.
Archive
Older pieces, kept online for reference.
No articles match that search. Try another word, or browse all topics.