OccupySaaS

OCCUPYSAAS.COM

OccupySaaS is an interesting concept, as explained on their website:

OccupySaaS is a protest movement written in code. Laid-off engineers, armed with AI and public documentation, are building clean-room, open-source, drop-in replacements for the platforms that rent our own industry back to us — one seat at a time.

Clean room rewrites are an interesting concept, relying on the nuances of copyright law, that ideas and functionality cannot be copyrighted, only their specific implementation. What this means is that you are legally permitted to clone an application if you create a specification, based on its observed behaviour, then hand that off to another team for re-implementation. However, as we saw earlier this year, AI can entirely automate this process, allowing for automated cloning and relicensing.

So far OccupySaaS has cloned Salesforce and Auth0. In a similar vein, ArtCraft have created Rust-based clones of an impressive 16 apps, including Photoshop, Word, PowerPoint and many other productivity tools.

It’s an interesting concept, but I don’t want a Salesforce or a Word clone, I want a fundamentally better version of those applications, with a user experience redesigned from the ground-up. However, that requires human effort and is thus much more costly!

The PHP2Shell Vulnerability Heralds Cybersecurity Hell

IEEE.ORG

Last month, the PHP Foundation published an update that fixed an out-of-memory error that could be exploited to gain root access, one of the most critical forms of vulnerability. Across the globe, people scrambled to apply updates, with many high-profile websites going offline for a period of time. The vulnerability was discovered by a cybersecurity research team, with the assistance of OpenAI’s models.

This article describes the strain this put open source maintainers under, as they rush to address a growing number of vulnerabilities. Elsewhere we have seen the median time to exploit, the time between a vulnerability being disclosed and attacks being detected in the wild, drop from years, to months, to practically zero.

This article speculates that things are going to get worse, and for a while I believe they will. We have a lot of technical debt that needs addressing.

## Scaling Git at OpenAI

OPENAI.COM

Through the use of AI we are collectively producing a lot more code than we used to. This is causing strain on the infrastructure that hosts our code and supports collaboration. GitHub has been experiencing reliability issues due to this increased activity, recently disclosing that commits had doubled in the space of just four months. The pace of AI is also causing issues within organisations that host their own source code, often within monorepos.

Within the frontier labs, they are creating code at a furious pace - the growth within OpenAI appears exponential:

openai commit growth

Just a couple of weeks ago Anthropic announced that they have done away with git completely, replacing it with a new version control system called Scion, which has a client-server architecture, in contrast to git’s distributed approach.

OpenAI have taken a different approach with Pando, their drop-in git replacement. They have taken an evolutionary approach, improving scalability by caching packfiles, sharding repos to allow horizontal scaling and employing weaker repository-wide consistency. It’s a shame that OpenAI haven’t open sourced Pando, but this isn’t all that surprising - open in name, closed in practice.

Hacker News community members share their coding setups

YCOMBINATOR.COM

A Hacker News community member asked “What’s your current AI agentic setup?” sparking a lively discussion with 678 comments. There’s a lot to digest there, which is why it might be simpler to read this digest from the AI News Report. As well as detailing the most popular tools (Claude Code wins, of course), this digest highlights some common usage patterns which are worth reviewing if you’re looking for ideas.