River's webstuff

As fun as AI is, we built so much bad software over the last 30 years, and now the universe is punishing us for our sins.

source

note : source redirects to simon willison's website, simon's website redirects to X, I have no interest in linking to X, given it's run by a facist.

Obviously, there is a junior developer crisis now and software engineers are being fired en-masse because we suddenly decided that AI can take over jobs now.

So here's the claim I am making - I think AI replaced the "old software engineering model", but nobody can use that anymore.

So, firstly, although we've had checks in the past for linting, test driven development, CI/CD, etc. The reality is that not every developer has been rigerous with the usage of these coding tools over time.

It's really interesting, because humans have been writing slop code for a very long time (or AI won't be writing sloppy code!), and we've been following bad software practices for a very long time, don't believe me?

Remember the old twitch source code leak, where the developers

1. AWS credentials hardcoded in source code.
2. Twitch client secrets and access tokens stored in a test file within the repository.
3. A private cryptographic key committed to the repository.
4. A Google API key hardcoded in source code.
5. A plaintext database password embedded in a Terraform configuration file.
6. BGP authentication passwords stored in network configuration files within the repository.

Or remember when the entire codebase for Black was one giant unsplit python file?

Or when Uber committed admin keys with admin priviledges directly to their source code repository, twice, once in 2014 and once in 2016

Or when equifax stored it's administrative credentials in plain text

In a world wherein agents can figure out serious software CVEs simply by the rumor of it, these lax practices are no longer going to be okay.

The mean time for discovering a bug is now -7 days, be damn sure when your source code leaks, there are probably a bunch of agents already trying to analyse it.

This creates a much stronger need of something else less documented, and that's responsible software, it's the instilling of paradigms within people, we are trying to replace this with "harnesses", we are failing, AI will write slop unless you really really ask it not to.

The future of software development is responsible software, it's developers who are aware of doing the right thing and as assholes about it. We can't go back to a time without AI, but we can sure be responsible against it.