Programming on Sumner Evans

How I Want to Use AI

Fri, 25 Sep 2026 06:25:00 -0600

TL;DR

The chat-based workflow encouraged by existing AI harnesses results in a fragmented source of truth for the desired functionality of the software.

To centralise the source of truth, we should use Markdown as Source (MaS). We should adopt a Specify, Generate, Inspect, Loop (SGIL) workflow which puts human-written Markdown files stored in source control front-and-centre.

My Current AI Workflow

I hate my AI workflow. I don’t know how everyone else is using AI, but currently my workflow looks something like:

Open Claude and type a chat message indicating what I want it to do.
Look at the code or plans it generates and test the functionality of the code it produced.
Type a chat message clarifying my ask or correcting the AI’s assumptions. Normally this involves telling it what I dislike about the code it generated.
Go to 2.

Forget MCP, Write CLI Apps

Tue, 02 Dec 2025 06:25:00 -0600

Over the last month or so, I’ve been using Claude Code for assisting with development tasks¹. One of the things that has confused me the most about the LLM-assisted programming ecosystem is the Model Context Protocol (MCP) landscape. In theory, MCP is a way to provide the LLM with data from an external system through a structured protocol. However, it sucks. It turns out that it is more efficient to have the AI just write code to analyse and manipulate the data from the external system directly. Rather than piping the data from the external system into the LLM’s context window and letting the LLM muddle through it, the LLM-written code can just do the necessary data processing. Computers have been good at running instructions on large datasets for a very long time and AI is not going to be a substitute for that.

Learning Typst

Sat, 11 Oct 2025 06:25:00 -0600

A few weeks ago, Typst came back on my radar because Sylvan Franklin’s videos on the topic came into my YouTube feed. In particular, Rewriting my resume in Typst in which he rewrites his resume from \(\LaTeX\) to Typst piqued my interest. I have maintained my resume in \(\LaTeX\) since I started at Mines in 2016, and so I figured this would be an opportunity for me to learn something new.

Vibe Coding Doesn't Require LLMs

Sun, 10 Aug 2025 06:25:00 -0600

I recently read I Know When You’re Vibe Coding by Alex Kondov in which he described certain characteristic “smells” of vibe-coded software. For example, he points to LLMs proclivity to pave their own path and go against established patterns within a project.

The article resonated with me, but I realized that it resonated with me at a deeper level than just explaining my interactions with LLM-written software. I realized that I’ve seen these “smells” before in pre-LLM software (especially that written by students).

Matrix Cryptographic Key Infrastructure

Sat, 21 Sep 2024 10:00:00 +0200

End-to-end encryption is one of the things which brought me to Matrix, and I’m sure that it’s one of the factors that brought many of you to Matrix as well.

However, Matrix’s user experience with cryptography is often confusing. I mainly blame the other chat networks for their incompetence. Most other chat networks don’t even provide any cryptographically-guaranteed security and privacy. Some networks provide encryption in a way that does not truly leave the user in control of their keys. Only a few networks (Signal) truly leave the user in control, and their UX is arguably worse than Matrix.

Message Security in Matrix

Fri, 09 Aug 2024 06:00:00 -0600

In Matrix, message security is provided by Megolm. The goal of message security is twofold. We (a) want messages to be encrypted on the sender’s device and (b) only be able to be decrypted on devices of users who are a part of the conversation. As a corollary, we want to prevent other parties (rogue homeservers, man-in-the-middle attackers, etc.) from decrypting the messages.

Matrix Cryptography Prerequisites

Fri, 09 Aug 2024 05:00:00 -0600

Understanding Matrix cryptography requires understanding some basic cryptography primitives. This article intends to explain those prerequisites in simple terms. This article is not a rigorous, mathematical description of the cryptosystems in use. Rather, it is a practical guide to what functionality each of the cryptography primitives provides. I’ve tried to include links to external resources if you are interested in learning more about a specific topic.

Differential Fuzzing Libolm and Goolm

Sat, 01 Jun 2024 06:25:00 -0600

Background

When we at Beeper started working on the new Android app, based on mautrix-go, we initially tried using goolm as the crypto implementation instead of libolm. We almost immediately ran into issues for unknown reasons, and rather than fighting with cryptography, we decided to just put in the work to get libolm working in the Android app.

Advent of Code 2023

Thu, 30 Nov 2023 09:00:00 -0600

Every year since 2015, Eric Wastl creates a two-part programming problem for each of the 25 days of Advent. He publishes a new problem every day at adventofcode.com at exactly midnight EST which is 22:00 the day before for me in MST. The last two years, I decided to do the problems as soon as they came out and streamed my problem solving sessions on my Twitch channel and uploaded them to my Youtube channel. The last two years I updated a blog post about each of the days, and I will try and do that this year as well.

Advent of Code 2022

Thu, 01 Dec 2022 09:00:00 -0600

Every year since 2015, Eric Wastl creates a two-part programming problem for each of the 25 days of Advent. He publishes a new problem every day at adventofcode.com at exactly midnight EST which is 22:00 the day before for me in MST. The last two years, I decided to do the problems as soon as they came out and streamed my problem solving sessions on my Twitch channel and uploaded them to my Youtube channel. Last year, I also updated a blog post after each day which I intend to do again this year.

Advent of Code 2021

Wed, 01 Dec 2021 09:00:00 -0600

Every year since 2015, Eric Wastl creates a two-part programming problem for each of the 25 days of Advent. He publishes a new problem every day at adventofcode.com at exactly midnight EST which is 22:00 the day before for me in MST. Last year, I decided to do the problems as soon as they came out and streamed my problem solving sessions on my Twitch channel and uploaded them to my Youtube channel.

Respecting Theme Preferences on Your Website

Tue, 07 Apr 2020 15:00:00 -0600

You may have noticed that dark themes are becoming more and more common across the computing landscape. Everything from Windows 10, macOS Mojave and later, iOS 13+, and Android 10+ to many Linux desktop environments and many individual browsers are including dark/light theme toggle settings.

Setting up Pelican to Automatically Deploy to GitLab Pages

Tue, 27 Aug 2019 23:58:00 -0600

Warning

This article is out of date, and may contain outdated information.

Since writing this article, I have made a few major shifts in my personal website infrastructure. I migrated from GitHub to GitLab and subsequently from GitLab to sourcehut. Then I migrated from Pelican to Hugo and hosted my website on a Linode VPS for a while before migrating back to GitHub and GitHub Pages.

I posted last year about my switch from WordPress to a statically generated site generated by Pelican by GitHub Pages in a blog post. Since then, my hosting situation has changed a couple of times. I first migrated to deploy straight to a DigitalOcean Droplet from Travis CI. Then, I migrated to GitLab and used GitLab CI/CD to do the same thing. Now that GitLab Pages has support for automatically handling the Let’s Encrypt SSL Certificate on custom domains, I’ve moved to GitLab Pages for hosting as well.

Setting up Pelican to Automatically Deploy to GitHub Pages

Mon, 11 Sep 2017 09:02:00 -0600

Warning

This article is out of date, and may contain outdated information.

I recently converted my site from WordPress to a statically generated Pelican site hosted on GitHub Pages. One of the most difficult parts of the transition was setting it up to automatically deploy to GitHub Pages. In this article, I will describe how I set this up.