Insights for great software & great teams

Dave Mosher explores using LLMs as digital archivists and resurrects his old GeoCities website. (Remember those!)

Davemo shares early experiments with Sora 2: discovered some interesting remix bugs and played around with applying it towards a game idea he's been noodling on.

A Slack conversation left Doc reflecting on what bug bashes say about the systems we work in.

Friday afternoon hot take (after a couple months of engineering exclusively with coding agents):

Context/prompting is a crutch that will only get you so far.

The AI agent does not have a true understanding of both:

• The business logic that the code it is generating represents
• The wider system(s) that it is being written to run within

IOW the why of the code.

The result is that no matter how hard you try it is not able to consistently apply rules and/or conventions (eg: logically organizing/co-locating code).

Test-driven development is to programming what double-entry bookkeeping is to accounting.

The fundamental character flaw of the software creator is not the misguided belief that they can solve any problem with software. Instead, it is the belief they can understand how their own software works before they create it.