HAL Update: Cache Control Plane Complete

The control plane for the cache is done. Code written. Test pass. Ship it right? Yea maybe, but then what?

As Charity Majors said on the Modern Software Engineering YouTube channel, the only thing that matters is production. Or to put it another way, what we trust – above all else – is battle tested code. I can’t think of anywhere this is truer than in HPC.

So, let’s say the code is perfect, the test coverage is absolute, and AI never makes a mistake. Hallucinations? Gone. Would you trust this code? Yea me neither. The question now becomes, what needs to happen to build the trust?

I think our level of trust in AI generated code is lower than that of Human created code but fundamentally the problem here would be the same either way. Trust and confidence in software is built over time and real world usage. Changes to that software are made iteratively over long periods of time not only because it takes time to create the software but also because it takes time (much more time) to build real world confidence in the software.

If we massively accelerate the rate at which software is created (and I genuinely think we are starting to see that), how do we accelerate adoption of that software. Specifically, I’m not talking about software with low downside risk like Moltbook. I’m talking about software that runs our banks and nuclear facilities.

I don’t have an answer to that. I’ve seen enough new schedulers and workload managers in HPC to know the problem was never writing a new one. Getting adoption though, that’s a different matter.

Anyway, the data control plane is done, I guess the next part is to see it build out the rest of the data plane.

Full prompt and change history in GitHub

GitHub - hmxlabs/hal

Contribute to hmxlabs/hal development by creating an account on GitHub.

GitHubhmxlabs