Emmz Rendle

Seventeen Years Late

Thu, 13 Aug 2026 00:00:00 +0000

Seventeen years ago I made myself ill working 100+ hours a week for months on end. So ill that I was in hospital for six weeks. Luckily for me, the company I was working for did provide private health insurance, so it was a very nice hospital. It was also summer, so when we weren’t in group therapy sessions, we were out in the sun, enjoying the gardens. I also made friends with several other patients and we would spend our spare time together, mostly just chatting about the groups and the world and our lives.

Effective Claude Code Workshop

Fri, 07 Aug 2026 00:00:00 +0000

Exciting news! I am running a new workshop, Effective Claude Code, in collaboration with Dylan Beattie. We’re doing an online version that is open to anyone, starting on the 1st and 2nd of September; it’s split over two days, from 2pm to 6pm UK time, so it’s accessible to people in all kinds of timezones and it doesn’t use up a whole work-day. We’ll probably try putting one on at a weekend so if you can’t get or afford the time off work (hello fellow freelancers) you can just sacrifice watching some sportsball or playing Elden Ring or whatever you do to unwind.

Maybe stop using Opus for everything

Wed, 05 Aug 2026 00:00:00 +0000

Saturday. I’m absent-mindedly baby-sitting a Claude Code session, applying a fairly simple change to an as-yet-unpublished greenfield project. And it’s taking aaaaaages. I’ve got an Opus 5 top-level session managing two Opus 5 agents, worker and reviewer, and between them they’ve been working on one task in one section of a change for almost a day and a half. And when I dig a little bit into why, it looks like one of them has decided that mutation testing is necessary for every change it makes (it’s not), and it’s running a dozen cycles of manual mutation tests on every test suite as part of every review, so what should be a two-minute “run tests and check code” turned into over an hour of generating sed scripts and whatever other nonsense. The manual part is because my global CLAUDE.md file says “don’t use JavaScript frameworks” and it extended that to mean “don’t use any packages that call themselves frameworks in any language ever”, so it decided it couldn’t use e.g. Stryker.

Educating Emmz

Tue, 04 Aug 2026 00:00:00 +0000

I went to see The Odyssey on Sunday. I loved it: I have never read the original source material so I don’t know what Christopher Nolan changed in order to to fit twelve thousand lines of dactylic hexameter verse¹ into a three-hour movie.

As I was doing some chores earlier in the day, it occurred to me that I didn’t even know when that epic poem was written, or when Homer lived relative to what I do know of Ancient Greece. So I opened Claude Desktop, selected Sonnet 5, and gave it this prompt:

The Claude extension I forgot I was using

Tue, 28 Jul 2026 00:00:00 +0000

A few months ago, while I was working on the memory layer for Daemons, a friend of mine who works at Yugabyte mentioned that they were working on something similar but at a much larger scale; I mean, it’s Yugabyte, massively distributed data storage is what they do. About a month later she invited me to join the early access program for Meko, which I did. I wrote a Daemon extension to work with it, and I added it as an MCP server for Claude Code. And then I kinda forgot about it for a couple of weeks (yayDHD), and then remembered it again when I collapsed a bunch of Daemon repos including dmon-meko into a monorepo, and wondered why I’d forgotten and figured it was because I hadn’t told Claude to use it.

Evolving Workflows

Thu, 09 Jul 2026 00:00:00 +0000

I just finished a session with Claude Opus, updating my plug-ins to reflect the changes I’ve made to my day-to-day workflow since publishing the first version.

What does that mean?

Okay, so when you’re working with Claude Code, there are lots of ways you can break down the work, and help the agent(s) to do their jobs more effectively. There’s the CLAUDE.md file, and skills, commands and sub-agents, and all of these work together to solidify your processes, standards and so on.

New Blog; Who Dis?

Wed, 08 Jul 2026 00:00:00 +0000

I appreciate that the post before this one says it’s the first one, even though there’s one “before” that. Both of those were placeholders generated by Claude Code while it was designing this theme. It came up with the Ink Console name too; it’s as good a name as any.

Enough of that, though. This post is written by me, and all future posts will also be written by me, unless otherwise specified. They’re going to be about AI, because what the hell else am I going to be writing about in 2026? AI, LLMs, agents, harnesses, runtimes, tools, frameworks, maybe hardware, and projects I’m working on with all of that. Ooh, that’s a good idea.

About

Tue, 07 Jul 2026 00:00:00 +0000

I’m Emmz Rendle. I write here about Claude Code, agentic development, and the workflows that make AI-assisted software reliable instead of chaotic.

This site is built with Hugo and the Ink Console theme — dark-first, self-hosted type (Fraunces, JetBrains Mono, Inter), and a single teal accent.

Elsewhere

Hello, Ink Console

Tue, 07 Jul 2026 00:00:00 +0000

Welcome to the new blog. It’s built with Hugo and a theme I’m calling Ink Console — dark-first, monospace where it counts, and quietly opinionated about type. This post is here mostly to show the theme off, so let’s exercise the elements.

Two voices, one accent

The design runs on two typefaces doing two jobs. Fraunces is the human voice — headings, the display title, anything that wants warmth. JetBrains Mono is the machine voice: labels, metadata, the little ▮ in the logo, and inline code like this. Inter stays out of the way for body text.

Spec-Driven Development, With Agents Doing the Typing

Sat, 20 Jun 2026 00:00:00 +0000

The fastest way to get unreliable code out of an AI agent is to give it a vague prompt and a lot of freedom. The fastest way to get reliable code is to give it a spec and almost none.

The loop

Everything I build with agents now flows through four steps:

Explore — think through the problem before writing anything down
Propose — capture the change as a spec: proposal, design, tasks
Apply — implement it section by section, each one reviewed
Archive — fold the shipped change back into the record

The agent types. I direct. The spec is the contract between us, and the review is where trust is earned.