zlacker

>>simonw+(OP)
With everything that we have done so far (our company) I believe by end of 2026 our software will be self improving all the time.

And no it is not AI slop and we don't vibe code. There are a lot of practical aspects of running software and maintaining / improving code that can be done well with AI if you have the right setup. It is hard to formulate what "right" looks like at this stage as we are still iterating on this as well.

However, in our own experiments we can clearly see dramatic increases in automation. I mean we have agents working overnight as we sleep and this is not even pushing the limits. We are now wrapping major changes that will allows us to run AI agents all the time as long as we can afford them.

I can even see most of these materialising in Q1 2026.

Fun times.

>>_pdp_+op1
What exactly are your agents doing overnight? I often hear folks talk about their agents running for long periods of time but rarely talk about the outcomes they're driving from those agents.