Augmented Coding Weekly

Issue #66

LGTM (Looks Good to Me)

YOUTUBE.COM

“Looks good to me” is a phrase that is often used, quite flippantly, when approving something that you haven’t really had time to read or review. A 10,000 line PR written by Claude? Sure … LGTM.

This AI-generated music video pokes fun at over-zealous AI adoption, and the ship-it (LGTM) culture. The song was created by Suno, and the video is a combination of TypeScript and three.js. Both are really impressive, and the song itself is worryingly catchy. I might even buy the T-Shirt!

The October inflection point for vibe game development

SIMONWILLISON.NET

Around one year ago we saw the ‘inflection point’ for coding agents, where AI moved from a tool that simply helped developers write code to autonomously doing meaningful software-engineering work. And the level of autonomy, and size of task, has grown rapidly since.

In this post Simon describes a new inflection point, this time for game development. He shares a number of games he has recently vibecoded (in the classic sense, no looking at the code), and the results are really impressive. Firstly, the coding agents now have a fantastic grasp of 3D rendering and game mechanics (Simon typically prompts them to use three.js specifically), but there is a more fundamental shift here - Simon claims that these models have a much better understanding of ‘fun’. He simply asked Claude to “Make it a LOT more fun” - and it did!

This reflects some of my own experiences with Opus 5.5 specifically. While I haven’t used it to create a game, I have found that its judgement and overall expertise has improved significantly. These models have been incredibly good at writing code for over a year, but now they are coupling it with engineering expertise, judgement, and dare I say it … taste.

The Mythical Man-Month in the age of AI

J11G.COM

The Mythical Man-Month is a book that was published way back in 1975, and is where the widely quoted Brooks’ Law emanated from:

“Adding manpower to a late software project makes it later.”

But does this hold up in the age of AI? Onboarding agents is rapid and cheap, so why can’t you just scale up your (agent) team in order to save a project that is behind? The problem is, the bottleneck moves elsewhere, from writing the code, to defining behaviours, or validating outputs.

Software may be getting worse, but it used to be horrible

BITLOG.COM

There is a growing argument that as software becomes cheap and easy to build, the overall quality is dropping. Put simply, people are shipping AI slop. While there are studies that indicate growth in code smells or issues due to AI generated code, I think most people are making a more emotional expression here, that software lacks polish, rather than has functional defects.

This blog post is a fun trip down memory lane (for those of us who are old enough), showing just how bad software used to be. And my god, it was truly awful! Blue screen of death, disk defrag, no security - it’s quite a list.

Modern software isn’t perfect, but it is so much better than the software many of us grew up with.

We should be more ambitious with the software we build

DANLUU.COM

Coding agents make it dramatically cheaper to build things, and as a result projects that we previously wouldn’t have considered because of the cost or time they would take, are now feasible. I flippantly tell people AI agents are “making the impossible, possible”. That migration project that would have taken 10 years to complete, and has had the budget request turned down each year, perhaps it is now possible and affordable?